# **The Canadian Information Ecosystem during the 2025 Federal Election**
October 2025
The Media Ecosystem Observatory is an interdisciplinary research initiative dedicated to analysing the complex web of online harms and digital threats of democracy, while actively working to safeguard against them.
The Observatory coordinates and supports the Canadian Digital Media Research Network. The CDMRN is a pioneering initiative committed to fortifying and fostering resilience within Canada's unique information ecosystem. Its mission is to understand the dynamics of information production, dissemination, and consumption across digital media with the goal of empowering Canadians to navigate the complexities of the modern digital age.
This project has been made possible in part by the Government of Canada. Ce projet a été rendu possible en partie grâce au gouvernement du Canada.
# **Contents**
| Executive summary
1 | | 3. Election integrity24 | |
|-----------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------|-----------------------------|-----------------------------------------------------------------------------------------------------------------------|
| Sommaire exécutif1 | | | 3.1. Foreign influence .
24 |
| 1. Introduction
5 | | 3.1.1. | United States .
27 |
| 1.1. Overview of the information ecosystem approach . 6 | | 3.1.2. | China .
28 |
| 1.2. Context going into the election .
6
2. Election information environment
8 | | 3.1.3.
3.1.4. | India 29
Russia 31 |
| | | 3.2. Misinformation .
32 | |
| 2.1. Survey .
2.1.1. | 8
Where did Canadians get | | 3.3. Bot and inauthentic activity .
32 |
| | their information? .
9 | | 3.4. Generative artificial intelligence .
32 |
| 2.1.2. | Who did Canadians trust
for information? .
9 | | 4. Information incidents
34 |
| 2.1.3. | What were the most important issues | | 4.1. Moderate incidents (level 2) .
35 |
| 2.1.4. | for Canadians? 10
How do Canadians discuss politics in | 4.1.1. | Fraudulent political news content
linked to cryptocurrency scams .
36 |
| | the context of growing polarization?
12 | 4.1.2. | Surge in claims undermining
election integrity 39 |
| 2.2. Digital Trace
2.2.1. | 13
Exploratory data analysis
14 | | Spotlight .
39 |
| 2.2.2. | Thematic analysis of candidates' social | | 4.2. Minor incidents (level 1) 44 |
| | media keywords and topic analysis of | 4.2.1. | Suspicious 51st state Facebook groups .
46 |
| 2.2.3. | the Canadian information ecosystem
16
Gender and political messaging
in Canada
16 | | 4.2.2. Suspect American influence in campaign
advocating for annexation of Alberta
as the 51st US state .
46 |
| 2.2.4. | Influencers linking media, politics,
and the public
19 | | 4.2.3. Suspicious U.Sbased recruitment
of Canadians .
47 |
| 2.3. Avatar-based social media monitoring
19 | | | 4.2.4. Suspicious amplification of |
| 2.3.1. | Dominant narratives .
20 | | anti-liberal narratives between
RT and right-wing outlets .
47 |
| 2.3.2. | Comparing platform discussions .
22 | 4.2.5. | Accusations of economic ties between |
| 2.3.3. | Comparing dominant actors .
22 | | Mark Carney and China .
48 |
| | 2.3.4. Engagement overview .
22 | | 4.2.6. Alleged Chinese pro-Liberal influence
on WeChat 48 |
| | 2.4. Incident response .
22 | | Spotlight .
50 |
| 2.5. Story Chaser and public tipline .
23 | | 4.2.7. | Alleged Chinese pro-Conservative
influence on RedNote
51 |
| | 4.2.8. Surge in faceless accounts and
deepfakes of politicians .
52 |
|--------------------|------------------------------------------------------------------------------------------------------------------------------------|
| | 4.2.9. High engagement in false claims of |
| | Mark Carney-Jeffrey Epstein relationship53
Spotlight .
53 |
| | 4.2.10. Inauthentic accounts on X sharing |
| | right-leaning election content 55 |
| | 4.2.11. Anti-Carney and Liberal Party
networks on X 55 |
| | 4.2.12. Liberals planting anti-Conservative buttons
and election integrity concerns .
55 |
| | 4.3. Potential incidents (level 0) .
56 |
| | 5. Findings and limitations57 |
| 5.1. Findings . | 57 |
| 5.1.1. | Influencers are now the loudest voices in the
online political information environment 57 |
| 5.1.2. | Generative AI—powered misinformation
emerged as a key challenge 57 |
| 5.1.3. | Political and platform segregation was
evident during the campaign 57 |
| 5.1.4. | Foreign interference was a central
public concern but interference attempts
appeared minor .
57 |
| 5.1.5. | Targeted manipulation attempts and high
volume of claims questioning election
integrity are undermining public trust .
58 |
| 5.2. Limitations . | 58 |
| 5.2.1. | Research, innovation and
operational constraints .
58 |
| 5.2.2. | Avatar-based monitoring challenges .
58 |
| | 5.2.3. Analytical challenges in measuring
influence of international voices 59 |
| | 5.2.4. Data constraints .
59 |
| | 5.2.5. Limits of open-source intelligence .
59 |
| | 6. Recommendations
60 |
|-----------------------|------------------------------------------------------------------------------|
| 7. Methodology | 62 |
| | 7.1. Incident response methodology 62 |
| | 7.2. Survey methodology 64 |
| | 7.3. Digital Trace and social media analysis .
65 |
| 7.3.1. | Seed list .
65 |
| 7.3.2. | Media crawlers .
66 |
| 7.3.3. | Topic modelling .
66 |
| | 7.3.4. Election fraud and foreign influence
claims identification .
66 |
| | 7.4. Avatar-based social media monitoring .
68 |
| 7.4.1. | Approach .
68 |
| 7.4.2. | Needs assessment .
68 |
| | 7.4.3. Avatar creation and summary .
68 |
| | 7.4.4. Avatar-based social media
monitoring team 70 |
| | 7.4.5. Account creation 70 |
| | 7.4.6. Data collection
71 |
| | 7.5. Story Chaser data collection and analysis tool 71 |
| 7.6. Public tipline . | 72 |
| 8. Glossary | 73 |
| | Contributors 75 |
| | Endnotes 76 |
# **Executive summary Sommaire exécutif**
This report presents the findings of our monitoring of the 2025 Canadian federal election.
Our central judgment is that **the integrity and outcome of the election were not compromised.** We found no evidence of material covert foreign interference. However, we assess that both the United States and China impacted the Canadian information and political environment
during the election. The election revealed a digital information ecosystem under mounting stress, shaped by influencers, polarized platforms, and new technological risks.
**Influencers are now the loudest voices in the online political information environment**. Amid the ban on news sharing on Meta and the reduced visibility of news on X, influencers, rather than traditional media, politicians, or political parties, commanded disproportionate attention and drove the circulation of political con-
tent across platforms. Their ability to blend entertainment and politics positioned them as powerful agenda setters in ways that can sometimes complicate accountability, transparency, and fact-based debate during elections.
Ce rapport présente les résultats de notre surveillance des élections fédérales canadiennes de 2025.
Nous retenons principalement que **l'intégrité et l'issue de l'élection n'ont pas été compromises.** Nous n'avons trouvé aucune preuve d'ingérence étrangère secrète et significative. Toutefois, nous estimons que les États-Unis et la Chine ont eu un impact sur l'environnement
informationnel et politique canadiens pendant les élections. L'élection a révélé un écosystème numérique de l'information soumis à une pression croissante, façonné par des influenceurs, des plateformes polarisées et de nouveaux risques technologiques.
**Les influenceurs sont désormais les voix les plus fortes dans l'environnement de l'information politique en ligne**. Dans le contexte du blocage des nouvelles par Meta et de la visibilité réduite des nouvelles sur X, les influenceurs, plutôt que les médias traditionnels, les politiciens ou les partis politiques, ont bénéficié d'une attention
disproportionnée et ont favorisé la circulation du contenu politique sur les différentes plateformes. Leur capacité à mêler divertissement et politique les a positionnés comme de puissants acteurs qui définissent l'ordre du jour, d'une manière qui peut parfois compliquer la responsabilité, la transparence et le débat fondé sur des faits pendant les élections.
**Generative AI—powered misinformation emerged as a key challenge**. We identified widespread use of AI-generated fake news, deepfakes, and automated bot activity. False content mimicking trusted Canadian outlets such as CBC and CTV spread widely, distorting political debate and confusing voters. These technologies
amplified existing misinformation dynamics, evolving quickly and demonstrating their capacity to undermine trust at scale. While the immediate impact on the election appeared to be low, our evidence suggests we are in the warning stage of an evolving information dynamic that could be catastrophic for public opinion, truth, and trust in future elections.
**La mésinformation générée par l'IA est apparue comme un défi majeur**. Nous avons constaté une utilisation généralisée des fausses nouvelles générées par l'IA, des hypertrucages et de l'activité des bots. De faux contenus imitant des médias canadiens de confiance tels que la CBC/ Radio-Canada et CTV ont été largement
diffusés, faussant le débat politique et semant la confusion dans l'esprit des électeurs. Ces technologies ont amplifié la dynamique de mésinformation existante, évoluant rapidement et démontrant leur capacité à saper la confiance à grande échelle. Bien que l'impact immédiat sur l'élection semble avoir été limité, nos données suggèrent que nous sommes dans la phase d'alerte d'une dynamique d'information évolutive qui pourrait être catastrophique pour l'opinion publique, la vérité et la confiance à l'égard des élections futures.
**Political and platform segregation was evident during the campaign**. Liberal-aligned voices and candidates continued to partially migrate to platforms such as Bluesky, while Conservative-aligned voices and candidates dominated X. TikTok use remained marginal for political candidates but proved effective for Canadian news
outlets and influencers. This digital segmentation risks hardening partisan and information divides and weakening the broader resilience of the Canadian public to misinformation and manipulation.
**La polarisation politique et la segmentation des plateformes étaient évidentes pendant la campagne**. Les voix et les candidats alignés avec le Parti libéral ont continué à migrer partiellement vers des plateformes comme Bluesky, tandis que les voix et les candidats alignés avec le Parti conservateur ont dominé X. L'utilisation de TikTok
est restée marginale pour les candidats politiques, mais s'est avérée efficace pour les médias d'information et les influenceurs canadiens. Cette segmentation numérique risque de renforcer les divisions partisanes et informationnelles, et d'affaiblir la résilience générale du public canadien à la mésinformation et à la manipulation.
**Foreign interference was a central public concern but interference attempts appeared minor**. We found only minor attempts by foreign actors to influence politics here in Canada through the information ecosystem. Their efforts achieved limited engagement. At the same time, U.S. politics, especially Trump, tariffs, and an-
nexation rhetoric, featured prominently in Canadian discourse. U.S. politics was a primary concern for many Canadians, and had a material impact on the voting intentions of millions of Canadians.
**L'ingérence étrangère était une préoccupation centrale du public, mais les tentatives d'ingérence semblaient mineures**.
Nous n'avons constaté que des tentatives mineures de la part d'acteurs étrangers d'influencer la politique au Canada par le biais de l'écosystème de l'information. Leurs efforts n'ont suscité qu'un engage-
ment limité. En parallèle, la politique américaine, en particulier la rhétorique de M. Trump, des tarifs douaniers et de l'annexion, a occupé une place prépondérante dans le discours canadien. La politique américaine était une préoccupation majeure pour de nombreux Canadiens et a eu un impact important sur les intentions de vote de millions de Canadiens.
**Targeted manipulation attempts and high volume of claims questioning election integrity are undermining public trust.** On
X, suspicious accounts targeted political leaders, most notably Mark Carney, as content moderation standards weakened and enforcement declined. In the latter half of the campaign, claims alleging sys-
temic election fraud gained traction online, especially on X and YouTube. This is consistent with an emerging pattern of election fraud claims surrounding democracies worldwide. While overall public trust in Canada's electoral process remains relatively strong, partisan divides on questions of election integrity are widening.
**Taken together,** these findings reveal an information environment shaped by influencers, digital manipulation and disruption, political polarization, and weakened platform governance and accountability. While these dynamics did not alter the outcome of the 2025 election, they are steadily eroding the foun-
*Without urgent reform, Canada is likely to enter future elections less prepared, more divided, and more exposed to manipulation.*
dations of democratic trust and resilience. The persistent spread of misinformation, the growth of partisan echo chambers, and the rise of unaccountable actors have created vulnerabilities that can be easily exploited. Most troubling is the absence of meaningful platform accountability and cooperation: Poor data access, weak enforcement, increased
lack of transparency, and systemic irresponsibility are limiting policymakers, researchers, and citizens' ability to respond effectively. Without urgent reform, Canada is likely to enter future elections less prepared, more divided, and more exposed to manipulation.
At its core, this project sought to protect Canadians from manipulation and dishonesty. While we recognize the profound limits to what we were able to observe, the dynamics and manipulations captured in this report present sufficient evidence to **sound the alarm**. Given the current state of data access, emerging
digital technologies, and the prevalence of malicious actors, Canadians are underprotected and vulnerable against manipulation.
**Les tentatives de manipulation ciblées et le grand nombre d'allégations mettant en cause l'intégrité des élections sapent la confiance du public**. Sur X, des comptes suspects ont ciblé des dirigeants politiques, notamment Mark Carney, alors que les normes de modération du contenu se sont affaiblies et que l'application de la loi a diminué. Au
cours de la seconde moitié de la campagne, les allégations de fraude électorale systémique ont gagné du terrain en ligne, en particulier sur X et YouTube. Cela correspond à une tendance émergente d'allégations de fraude électorale dans les démocraties du monde entier. Alors que la confiance du public dans le processus électoral canadien reste relativement forte, les clivages partisans sur les questions d'intégrité des élections se creusent.
**Dans l'ensemble,** ces résultats révèlent un environnement d'information façonné par les influenceurs, la manipulation et la perturbation numériques, la polarisation politique ainsi que l'affaiblissement de la gouvernance et de la reddition de comptes des plateformes. Bien que ces dynamiques n'aient pas modifié
*Sans une réforme urgente, le Canada risque d'aborder les futures élections moins bien préparé, plus divisé et plus exposé à la manipulation.*
le résultat de l'élection de 2025, elles érodent progressivement les fondements de la confiance et de la résilience de la démocratie. La diffusion persistante de mésinformation, la croissance des chambres d'écho partisanes et la montée en puissance d'acteurs non responsables ont créé des vulnérabilités qui peuvent être facilement exploitées. Le plus
inquiétant est l'absence de responsabilité et de coopération significatives au niveau des plateformes : L'accès insuffisant aux données, la faiblesse dans l'application de la législation, le manque croissant de transparence et l'irresponsabilité systémique limitent la capacité des décideurs politiques, des chercheurs et des citoyens à réagir efficacement. Sans une réforme urgente, le Canada risque d'aborder les futures élections moins bien préparé, plus divisé et plus exposé à la manipulation.
Au fond, ce projet visait à protéger les Canadiens contre la manipulation et la tromperie. Tout en reconnaissant les limites profondes de ce que nous avons pu observer, la dynamique et les manipulations décrites dans ce rapport constituent des preuves suffisantes pour **sonner l'alarme**. Compte tenu de l'état actuel de l'accès aux
données, des technologies numériques émergentes et de la prévalence des acteurs malveillants, les Canadiens sont sousprotégés et vulnérables à la manipulation.
### **We recommend immediate action.**
**Governments must** mandate transparency, empower regulators, enforce oversight, and address new challenges with AI-generated content.
**Platforms must** guarantee ad transparency, strengthen moderation, grant data access, and reinvest in information and election integrity initiatives.
**Traditional media** must ensure newsroom-level training and commitment to covering information manipulation, build trust through accessible and accurate reporting, and focus on holding other actors accountable.
**Influencers must** embrace democratic responsibility by being transparent, fact-checking, and fostering respectful civic dialogue.
**Civil society and researchers must** anticipate threats, expand civic education, and coordinate efforts to build democratic resilience.
**The public must** intentionally broaden their information environment, practice informed vigilance, and engage across divides to build community and resist manipulation.
### **Nous recommandons d'agir immédiatement.**
**Les gouvernements doivent** imposer la transparence, donner aux régulateurs les moyens d'agir, mettre en œuvre la surveillance et relever de nouveaux défis posés par le contenu généré par l'IA.
**Les plateformes doivent** garantir la transparence des publicités, renforcer la modération, permettre l'accès aux données et réinvestir dans des initiatives d'information et d'intégrité électorale.
**Les médias traditionnels** doivent assurer une formation au niveau de la salle de nouvelles et s'engager à couvrir la manipulation de l'information, à instaurer la confiance grâce à des reportages accessibles et rigoureux, et à se concentrer sur la responsabilisation des autres acteurs.
**Les influenceurs doivent** assumer leur responsabilité démocratique en faisant preuve de transparence, en vérifiant les faits et en encourageant un dialogue civique respectueux.
**La société civile et les chercheurs doivent** anticiper les menaces, développer l'éducation civique et coordonner les efforts pour renforcer la résilience démocratique.
**Le public doit** délibérément élargir son environnement d'information, pratiquer une vigilance éclairée et s'engager au-delà des clivages pour renforcer la communauté et résister à la manipulation.
**Anything less will leave Canadian democracy, stability, and prosperity ever-more vulnerable.**
**À défaut, la démocratie, la stabilité et la prospérité du Canada seront encore plus vulnérables.**
# **1. Introduction**
In recent years, the increasing fragmentation and complexity of Canada's digital information landscape have posed significant risks to public trust and democratic engagement. Addressing these challenges demands a coordinated, data-driven, and resilient response across disciplines and institutions. This document explores how three innovative, interconnected initiatives are rising to meet that need: the Media Ecosystem Observatory (MEO), the Canadian Digital Media Research Network (CDMRN), and the Coalition for Information Ecosystem Resilience.
The MEO is an interdisciplinary research initiative co-led by McGill University and the University of Toronto. Since 2019, the MEO has been pioneering an evidence-based model for assessing the health of Canada's information ecosystem — tracking online harms, disinformation threats, and structural vulnerabilities to democracy. The MEO uses advanced statistical and computational methods to study how information circulates, influences behavior, and impacts democratic discourse.
The CDMRN, launched in June 2023, is a pioneering national-level research community coordinated by the MEO. Funded by Heritage Canada's Digital Citizen Initiative, the CDMRN is dedicated to fortifying the resilience of Canada's information ecosystem by studying how content is produced, shared, and consumed online as well as empowering Canadians to make informed media choices. It empowers Canadian researchers to study ecosystem vulnerabilities, threats and cultivate ecosystem resilience.
Complementing these efforts is the [Coalition for Information](https://www.cdmrn.ca/coalition-for-information-ecosystem-resilience) [Ecosystem Resilience](https://www.cdmrn.ca/coalition-for-information-ecosystem-resilience), a network of researchers and civil society organizations collectively driven to safeguard the Canadian information ecosystem during moments of stress and/or instability. The coalition works together to identify threats, coordinate responses, and build public awareness and resilience in the face of information disruption. They actively support rapid incident response, amplify trusted content, and promote transparency and media literacy across communities.
The synergy created between these three initiatives enabled an unprecedented ability to understand, monitor, and safeguard the public domain of Canada's information ecosystem during the election. While this report is principally authored by the MEO team, this work would not have been possible without extensive knowledge sharing, ongoing communications and collective investigation with our partners. In complement to this report, our partners have also published many insightful reports, tools and findings and hosted events during the election period. This work was amplified in our MEO [Weekly Updates](https://www.cdmrn.ca/weekly-updates) for key partners including [Apathy is Boring](https://www.apathyisboring.com/), [PolCommTech](https://www.polcommtech.com/), [DisinfoWatch](https://disinfowatch.org/), [DFRLab](https://dfrlab.org/), [Information Integrity Lab](https://infolab.uottawa.ca/), [Digital](https://www.digitalpublicsquare.org/) [Public Square](https://www.digitalpublicsquare.org/), [Democratic Engagement Exchange,](https://www.engagedemocracy.ca/) [Samara](https://www.samaracentre.ca/) [Centre for Democracy](https://www.samaracentre.ca/), [Applied AI Institute](https://www.concordia.ca/research/applied-ai-institute.html), [The Centre for](https://democracy.ubc.ca/) [the Study of Democratic Institution](https://democracy.ubc.ca/)s, [Media Smarts](https://mediasmarts.ca/), and [Reset.Tech](https://www.reset.tech/).
The report consists of this introductory chapter, which provides an overview of the information ecosystem approach and the electoral context in which the 2025 Canadian federal election took place. [Section Two](#page-11-1) provides a data-driven view of the election information environment drawing upon survey, digital trace, and manual social media monitoring and reports on the incident response and tipline approach. [Section Three](#page-27-1) evaluates the main threats to election integrity including foreign interference, misinformation, bots and inauthentic activity and generative artificial intelligence. [Section Four](#page-37-1) provides an incident-oriented view of the election and describes each incident identified and investigated during the election. [Section](#page-60-1) [Five](#page-60-1) provides the main findings and limitations. [Section Six](#page-63-1) is the final section of the report and provides recommendations and conclusions. Detailed [Methodology](#page-65-1) and [Glossary](#page-76-1) sections are also included.
### **1.1. Overview of the information ecosystem approach**
The concept of an information ecosystem describes the complex interplay between media, technology platforms, political actors, influencers, and the general public, emphasizing how information is produced, disseminated, consumed, and influenced. In the context of the 2025 Canadian federal election, we employed a mixed-method approach to map and understand this ecosystem:
**Survey research:** Conducting comprehensive, repeated surveys of Canadians across the pre-election, campaign, and post-election periods to capture shifts in public trust, issue priorities, perceptions of misinformation and information threats (e.g., generative AI, foreign interference), and electoral integrity.
**Digital trace analysis:** Collecting and analyzing social media data from influential Canadian entities (politicians, media outlets, influencers, civil society groups) to track online discourse, engagement patterns, and information dissemination across multiple platforms.
**Avatar**1 **-based social media monitoring:** Utilizing controlled, avatar-based observations across social media platforms to capture real-time dynamics, identify emergent misinformation claims and narratives, and detect coordinated or inorganic activity. Avatars are strategically developed to encompass varied demographics, political partisanship, and priority issues and preferences.
**Incident response framework**: Applying a structured methodology for rapidly detecting, assessing, and responding to significant information-related threats and incidents during the electoral period.
**Crowdsourced reporting:** Engaging the public and civil society in reporting suspected digital threats through an integrated tipline system, enabling real-time detection and triage of misinformation, disinformation, and emergent online harms. This crowdsourced approach leverages a coalition of researchers and practitioners to analyze and escalate reports, enhancing the broader incident response framework and situational awareness during the electoral period.
Using this comprehensive approach, this study provides nuanced insights into the strengths, vulnerabilities, and overall health and resilience of the electoral information ecosystem in Canada, informing strategies to safeguard future elections and improve electoral integrity initiatives.
# **1.2. Context going into the election**
The 2025 federal election kicked off in a chaotic political context, with the resignation of Prime Minister Justin Trudeau and a subsequent Liberal leadership contest, heightened economic tensions with the United States, and a fraught information environment.
On Jan. 6, 2025, Trudeau prorogued Parliament and announced his intention to resign as Prime Minister following the election of a new leader for the Liberal Party of Canada. Former Governor of the Bank of Canada and the Bank of England, Carney won the subsequent party leadership race on March 9 and was sworn in as prime minister on March 14. On March 23, Parliament was dissolved and a federal election was called to take place on April 28.
At the time of Trudeau's resignation, Pierre Poilievre and the Conservative Party of Canada were anticipated to win the upcoming election, with [a lead of more than 20 percentage](https://338canada.com/polls.htm) [points in vote intentions](https://338canada.com/polls.htm). However, escalating tensions with the United States shifted the political conversation from discontent with the previous Trudeau government to widespread concern about how to [handle](https://www.bbc.co.uk/news/articles/cd7eyz3yn5do) relations with the United States under the newly elected U.S. President Donald Trump. In November 2024, Trump announced plans for 25% tariffs on all imports from Canada (and Mexico) and soon began referring to Canada as the future "[51st state](https://globalnews.ca/news/10923704/trump-canada-51st-state/)." When Trump's tariffs took effect on March 4, 2025, Canada responded with retaliatory tariffs and widespread action including Canadians and Canadian retailers [buying](https://globalnews.ca/news/11106170/buy-canadian-us-companies-impact-canada-retailers/) fewer American products, pulling U.S. [alcohol](https://globalnews.ca/news/10993794/trump-tariffs-liquor-boards/) from liquor stores entirely, and [cancelling](https://www.cbc.ca/news/canada/ottawa/angered-by-trump-s-threats-canadians-cancel-u-s-trips-1.7456243) travel plans to the States. The trade dispute with the U.S. quickly took on larger cultural and political significance for many Canadians (to the point that a hockey game between the U.S. and Canada held [geopolitical](https://www.cbc.ca/sports/hockey/4-nations-face-off-final-canada-united-states-1.7464559) weight) and defined much of the context of the federal election.
The election was further punctuated by rising concerns about the integrity of the Canadian information environment. The final report of a public inquiry into foreign interference, released in January 2025, [found](https://foreigninterferencecommission.ca/fileadmin/report_volume_1.pdf) that while foreign interference "may have had some impact" on previous elections, there is no evidence to suggest "that our institutions have been seriously affected by such interference." Rather, it asserted that the biggest [threat](https://www.cbc.ca/news/politics/final-report-public-inquiry-foregin-interference-1.7443597) to Canadian democracy was widespread information manipulation, disinformation, and misinformation. The 2025 election was also the first federal contest to take place following the rise of [generative AI](https://www.cpac.ca/articles/potential-misuse-of-ai-during-an-election) tools and Meta's 2023 decision to end [news](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada) availability for Canadian users on Facebook and Instagram, posing significant and novel risks to the Canadian media ecosystem.
Prior to the federal election, our team conducted a comprehensive environmental scan of the Canadian information and media ecosystem to identify threats and vulnerabilities to the election that would require close monitoring during the electoral period. To do so, we scanned information sources (using a mix of digital trace efforts, surveys, news analysis, and in-depth qualitative research) to identify potential areas of concern.
While our scan yielded a variety of threats to watch for during the election period, we identified three threats to be the most significant. First, election security. We suspected that the legitimacy of election results may be questioned, relating to controversy over fraudulent processes and interference with election technologies. We identified that these concerns were already [prevalent during the 2021 Canadian Election](https://osf.io/preprints/osf/ubfmx) and garnered substantive attention during the [2024 election in British](https://meo.ca/work/british-columbia-election-information-ecosystem-project) [Columbia](https://meo.ca/work/british-columbia-election-information-ecosystem-project). While provinces have established strategic plans to address the security of election technologies in recent years, public concern over effective measures to counter election interference persists. Second, foreign interference. Election periods, in general, provide a unique opportunity for foreign influence and interference. The threat of foreign interference has been [a persistent concern for both the Canadian public](https://foreigninterferencecommission.ca/fileadmin/report_volume_1.pdf) [and the governmen](https://foreigninterferencecommission.ca/fileadmin/report_volume_1.pdf)t, particularly from [China, Russia, India,](https://www.canada.ca/en/security-intelligence-service/corporate/publications/csis-public-report-2024.html) [Pakistan, and Iran](https://www.canada.ca/en/security-intelligence-service/corporate/publications/csis-public-report-2024.html). In the lead up to the election, public concerns over American influence dramatically increased. The uncertain and consistently changing relationship between the U.S. and Canada suggested increased vulnerability to information manipulation, disinformation, and inaccurate reporting. Finally, generative AI, like deepfakes and fake voting
information, posed a significant threat, capable of spreading disinformation and undermining public trust. AI tools can be exploited by foreign actors, political groups, or influencers to amplify false narratives, impersonate officials, and manipulate voter perception at scale. These threats are compounded by existing vulnerabilities in the information ecosystem, including polarization, misinformation on contentious issues, and gaps in electoral preparedness.
Ecosystem vulnerabilities spanned sensitive communities to contentious topics. Of particular concern, we identified that diverse diaspora communities were substantial populations and influential voters in Canada who could be key targets of foreign influence and transnational repression. In particular, Chinese and Indian communities in Canada are key targets of foreign interference and coordination disinformation campaigns to silence dissidents, suppress or sway voter decisions, and slow discord. Further, two topics of interest we thought likely to be targets of disinformation and information manipulation were immigration and climate change. Skepticism towards immigration is influenced by the affordability and housing crisis in recent years. This topic is susceptible to negative and misinformed claims around immigrants and racialized minorities as part of electoral campaign/policy discourse. The topic of climate change also intersects with ongoing discourses on the economy and extractive practices in Canada, strengthened by the lobbying efforts of the oil and gas industry. Disinformation around the impact of climate change may emerge in-hand with discussions on potential economic projects tied to oil and gas.
Conservative Party Leader Pierre Poilievre speaks at a rally in Penticton, British Columbia, April 5, 2025.
# **2. Election information environment**
The 2025 election took place in a rapidly evolving information environment marked by several transformative developments. Most notably, as Canadians increasingly consume news online and Facebook remains the most used social media platform in Canada, Meta's decision to ban news link sharing on its platforms in August 2023 has [significantly reduced \(incidental\)](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada) [exposure to news](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada) online and created an information void that has, in part, been filled by less trustworthy content. Elon Musk's acquisition of Twitter, which he rebranded to X, his political stances, and recent changes made to the platform have also contributed to greater fragmentation of the online information environment, as some centrist and left-leaning users [migrated](https://www.cbc.ca/news/politics/canadian-politics-bluesky-x-1.7391832) to alternative platforms such as Bluesky. While Canada has had a history of [high trust in mainstream media](https://ppforum.ca/articles/lessons-in-resilience-canadas-digital-media-ecosystem-and-the-2019-election/), [trust has](https://www.cem.ulaval.ca/wp-content/uploads/2025/06/dnrcanada2025e.pdf) [been declining in recent years](https://www.cem.ulaval.ca/wp-content/uploads/2025/06/dnrcanada2025e.pdf), further complicating efforts to maintain a shared factual understanding informed by rigorous journalism.
This section provides a description of the information environment during the election as seen by each subteam involved in the project, including [Survey](#page-11-2), [Digital Trace,](#page-16-1) [Avatar-Based Social](#page-22-1) [Media Monitoring](#page-22-1), [Incident Response](#page-25-1), as well as the [Story Chaser](#page-26-1) [and Public Tipline](#page-26-1). Specifically, the Survey section provides an overview of where Canadians were getting their information during the campaign, how much trust they had in different information sources, what issues they were the most concerned about, and their willingness to discuss politics online and offline in the context of growing affective polarization. In the [Digital Trace section](#page-16-1), we evaluate political actors' use of social media and online discourses during the election. We specifically analyze how much engagement they got across platforms, the topics they discussed, and their role within the broader information ecosystem. The [Avatar-Based Social Media Monitoring](#page-22-1) section highlights how the content encountered by our social media avatars changed across platforms and based on their socio-demographic and political characteristics.
Key findings from this section include that online political engagement surged during the election period, with interactions around election-related content rising by nearly half compared to the previous year. Canadians' attention, however, fragmented across platforms: TikTok became a powerful amplifier of news content, Instagram provided politicians with an effective space for visibility and interaction, and X remained the dominant hub for influencers, who played an outsized role in shaping narratives. This platform-specific configuration underscores that political communication in Canada is increasingly fragmented, more reliant on influencers than institutional actors, and deeply tied to the unique affordances of each platform. At the same time, surveys highlighted widespread public anxiety over misinformation, foreign interference, and the novel risks posed by generative AI. These concerns reinforce the sense of an ecosystem where engagement is intensifying, authority is shifting, and confidence in the reliability of political information is strained.
# **2.1. Survey**
The survey component of our project provides a critical window into how Canadians experienced the 2025 federal election. By capturing public opinion before, during, and after the campaign, these surveys allow us to trace not only where people sought political information, but also how they evaluated its credibility, which issues shaped their daily concerns, and how comfortable they felt engaging in political discussion. Because partisanship remains one of the strongest predictors of political attitudes and behaviours, our analysis focuses on Canadians who identify with a political party. This perspective enables us to understand how competing partisan communities navigated an increasingly fragmented and polarized information environment.
**Social media platform use for general and political purposes by partisan identification.**
#### **Figure 1. General and political use of social media platforms by partisan identity**
In the run up to the election, we fielded monthly tracking surveys of 1,400-1,500 respondents at the end of each month throughout 2024 and in January and February of 2025. Throughout the campaign period, from March 23 to April 27, we surveyed 200 Canadians per day, with approximately 7,000 respondents during the campaign week. Lastly, we followed up two weeks after the election campaign to recontact over 2,000 respondents from the campaign period to again measure their opinions to track changes over time.
The analysis in this section focuses on partisans: Canadians who express a clear preference for a political party. This emphasis is important for two reasons. First, election campaigns are competitions between political parties and provide the primary structure for structuring political life. Second, partisanship remains one of the strongest predictors of political attitudes and behaviours and can help us understand the overall election information environment.
To begin, we examine the foundations of the election information environment: where Canadians turned for both general and political content.
#### **2.1.1. Where did Canadians get their information?**
[Figure 1](#page-12-1) shows platform use for general and political purposes among partisans from the highest performing four parties. Across these parties, Facebook and YouTube were the most commonly used platforms for both generic and political information
sources, with Instagram closely following in third. TikTok and X are used less frequently than the previously mentioned three major platforms, though a notably larger share of X's users also use it for political purposes compared to any other platform, save for Bluesky. Bluesky remains marginal across all groups, although Liberals and NDP partisans see double the rates of self-reported use compared to Conservatives. Conservatives use X more than their counterparts, in both political and general use. These trends provide some evidence of an ideological preference of social media use, where supporters of right-wing parties are more active on X and supporters of centrist or leftwing parties are drifting to alternative options like Bluesky. Bloc Québécois supporters tend to be as active on Facebook but less active across other platforms.
While platform use reveals where partisans encountered information, it is equally important to consider how much confidence they placed in those sources.
#### **2.1.2. Who did Canadians trust for information?**
Compared to other information sources such as social media and podcasts, however, news outlets (national, local, and international) still commanded the highest overall levels of trust across the electorate. [Figure 2](#page-13-1) shows that Conservative partisans are much less likely to trust information about politics and current affairs from news websites in contrast to Liberal, NDP, and Bloc Québécois supporters. At the national level, for example, a little under half of Conservatives trust national news
**Figure 2. Trust in information relating to politics and current affairs from different sources by partisan identification.**
(48%) versus more than three quarters of Liberals (84%), NDP supporters (78%), and Bloc supporters (81%). The trend is similar for news at the local level, with 54% (Conservative) versus 79% (Liberal), 75% (NDP), and 82% (Bloc) and at the international level with 46% (Conservative) versus 67% (Liberal), 64% (NDP), and 70% (Bloc). In contrast, overall trust in digital-first sources remained much lower, with only 20 to 34% of partisans saying they trusted social media or podcasts. Yet this lower trust should be understood alongside usage patterns: Conservatives in particular were more likely than other partisan groups to use platforms such as X for political purposes, suggesting that even with skepticism about credibility, these digital spaces played a disproportionately important role in their information environment.
Beyond questions of access and trust, elections are also shaped by the issues that dominate public attention.
#### **2.1.3. What were the most important issues for Canadians?**
Donald Trump and the economy emerged as the dominant concerns for both Liberal and Conservative supporters throughout the 2025 election campaign. Among Liberals, NDP, and Bloc Québécois supporters, worry about Trump and potential tariffs consistently outpaced all other issues, remaining the top daily concern. For Conservatives, however, attention was split more evenly between Trump and the economy, as shown in figure 3, with the two issues competing closely for the lead. Other topics, such as health care, misinformation, the environment, housing, and AI in daily life, showed more moderate and relatively stable levels of concern across both groups. Notably, environmental concern remained markedly higher among Liberals than Conservatives, underscoring a persistent partisan divide on this issue.
[Figure 4](#page-14-0) shows that concerns about misinformation, artificial intelligence, and foreign interference cut across partisan lines, though important differences remain. Liberal and NDP supporters consistently report the highest levels of concern, with around two-thirds describing themselves as 'very' or 'extremely' concerned about misinformation (59% and 64%), AI and deepfakes (67% and 71%), and foreign interference (58% each). Bloc Québécois supporters also express high levels of worry, particularly about AI and deepfakes (67%), though they are somewhat less concerned about misinformation (60%) and foreign interference (55%). Conservatives are similarly concerned about Foreign interference, making this a rare cross-partisan issue of concern, but are less concerned about misinformation and AI/deepfakes. Taken together, these results highlight a broad consensus across parties that emerging technologies and disinformation pose serious challenges, but they also reveal partisan divides in the intensity of those concerns, with Conservatives notably more
**Figure 3. Percentage of Canadians reporting thinking daily about about different issues throughout the election campaign by partisan identification from March 24 to April 28, 2025.**
**Figure 4. Percentage of Canadians 'very' or 'extremely' concerned about different information manipulation challenges throughout the election campaign by partisan identification.**
**Figure 5. Feelings towards respondents' own party and other parties by partisan identification.**
skeptical about the threats of misinformation in particular.
These issue concerns unfolded within a highly polarized environment, which influenced not only how Canadians felt about opposing parties, but also how willing they were to discuss politics with others.
#### **2.1.4. How do Canadians discuss politics in the context of growing polarization?**
It is now a well-known fact that [Canadians are becoming in](https://doi.org/10.1017/S0008423923000112)[creasingly affectively polarized](https://doi.org/10.1017/S0008423923000112) (politically polarized) since the beginning of the 21st century. The inter-party sentiment matrix in [Figure 5](#page-15-1) highlights the intensity of affective polarization during the 2025 Canadian federal election. Each square corresponds to a pairing between supporters of a party along the left and sentiment towards the party along the top. For example, the top-left corner is how Liberal partisans view the Liberal Party. Supporters of both the Liberals and Conservatives expressed overwhelmingly positive feelings toward their own parties (average ratings of 79 and 83, respectively), but hostility toward one another. Liberal supporters rated Conservatives at just 25 on average, while Conservative supporters rated Liberals even lower at 22 (a score of 20 is extremely negative). Liberal partisans have modest feelings towards the NDP (51) and the Greens (44), but (unlike the other parties) Conservative Party
supporters strongly dislike all other political parties. Importantly, this polarization extends beyond the two main parties: NDP and Green supporters show moderately positive feelings toward Liberals but highly negative evaluations of Conservatives, while People's Party of Canada (PPC) supporters display the opposite pattern: warmth toward Conservatives but strong hostility toward Liberals, NDP, and Greens. Bloc Québécois supporters, meanwhile, maintain positive in-group feelings but relatively cool ratings toward all other parties. Taken together, the results illustrate a fragmented partisan landscape where cross-party sympathy is limited and often replaced by outright hostility.
In this context, we evaluated Canadians' willingness to engage in dialogue with one another. [Figure 6](#page-16-2) shows that about half of Canadians feel comfortable openly expressing their political views at work or at school. However, this number drops to just over one third when it comes to expressing those views online (including directly responding to online news content and social media posts that people may agree with). This pattern remained stable over the six months before the election and did not significantly vary during the election. This suggests that the political content encountered on social media is likely to disproportionately come from and reflect the views of the minority of Canadians who feel comfortable sharing their opinions in online spaces.
**Figure 6. Comfort expressing political views in different contexts from Sept 2024 to April 2025.**
While the survey highlights the perspectives and behaviours of citizens, we now shift to digital trace analysis to understand political actors themselves, exploring how parties, candidates, and influencers shaped the online information ecosystem.
# **2.2. Digital Trace**
By examining the online behaviour of parties, candidates, media outlets, and influencers, we can better understand how the information environment was shaped by those actively producing and disseminating political content.
Between February 23 and May 28, the Digital Trace team systematically tracked the activity of 4,009 Canadian entities (including politicians, media organizations, government bodies, civil society groups, and prominent online personalities) across six major platforms: X, Instagram, TikTok, Bluesky, YouTube, and Telegram.2 In total, this effort captured 1.52 million posts, providing a rich dataset to analyze the strategies, narratives, and dynamics that structured political communication during the election period. The methodology of identifying entities can be found in the [Methodology](#page-65-1) section.
This digital perspective complements the survey findings by allowing us to explore the "supply side" of the election information environment: What kinds of messages political actors sought to amplify, which platforms they prioritized, and how partisan and institutional divides shaped the flow of information online. Together, the survey and digital trace analysis offer a more complete picture of how both citizens and political elites engaged with the online political sphere in Canada's 2025 federal election.
Key findings from this section show that Canada's online political sphere during the 2025 election was both highly active and deeply uneven. Engagement surged nearly 50% compared to the previous year (see [figure 7](#page-17-1)), but was concentrated around a small number of politicians, influencers, and media outlets. Major party leaders remained central to online political communication, though right-wing influencers and infotainment outlets also played an outsized role in shaping discourse. Conservative politicians overwhelmingly prioritized X, while Liberals and New Democrats pursued broader multi-platform strategies, and TikTok emerged as a key site of news engagement despite limited posting by candidates. Thematic analysis revealed a fragmented agenda: Economic issues dominated candidate messaging, influencers drove partisan critiques and U.S.-related stories, and news outlets focused comparatively more on policy debates and international conflicts such as Israel-Palestine. Gender patterns in political messaging highlighted enduring divides, with male candidates more focused on security and foreign policy whereas female candidates emphasized caregiving
**Figure 7. Annual comparison of daily engagement (likes) with all seedlist content across six platforms from March 2 to May 25th in 2024 and 2025.**
and community while having equivalent audience engagement across genders. Finally, semantic network analysis demonstrated that influencers, more than parties or news media, acted as bridges across partisan clusters, diffusing narratives across otherwise siloed communities. Together, these findings underscore that Canada's digital campaign environment is increasingly shaped by a small set of high-visibility actors, platform-specific dynamics, and the central and mediating role of influencers.
#### **2.2.1. Exploratory data analysis**
The Canadian social media ecosystem was highly active during the election period; compared to the same period in the previous year, there was an increase of 47% in terms of engagement (the sum of likes, shares, and comments on posts from the influential entities). Politician entities unsurprisingly saw the greatest growth in engagement at the start of the election period as well as the steepest drop immediately after election day (April 28).
Across all types of entities studied (politicians, influencers, and news outlets), we found that the majority of the posting activity happens on X. Politicians post 69% of their posts on X, while X makes up 87% and 89% of posts from news outlets and influencers, respectively. Note that Canadian news organizations could not post their content on Facebook and Instagram during the election and links to their content could also not be posted. However, this does not always correspond with the platforms where the entities receive the most engagement. News outlets received the most engagement on TikTok (69%) while politicians received the most engagement on Instagram (48%). Only influencers received the most engagement on X (59%). This illustrates how Canadians use diverse platforms to engage with political and election content.
**Engagement with news, politicians, and influencers**: Our data shows that social media engagement is concentrated around a few entities. The top news outlets include major national outlets (CBC, CTV, and Global News) as well as online infotainment organizations (Narcity, Now Toronto). The top influencers are composed of mostly right-wing figures and groups (Canada Proud, Canadian Patriot). The top politicians are present and former party leaders (Poilievre, Trudeau, Carney, Jagmeet Singh) and provincial premiers (Wab Kinew of Manitoba). Politicians have a high degree of engagement concentration, with the top five politicians receiving 60% of all politician engagement. This is greater than the engagement received by the top five news (45%)3 and the top five influencers (22%). Canadians still largely look to Party leaders for the distribution of key political information and to show their support.
Figure 8.
Daily number of posts by major federal party candidates by platform.
#### Platform usage by daily post counts by candidates
Post count by platform across major federal parties: five-day rolling average
N = 198,123 posts from candidates of major parties.
There were striking differences in how candidates used social media by party, as shown in Figure 8. Conservative candidates concentrated their efforts overwhelmingly on X, often generating more than 600 posts per day at peak periods, even though Conservative candidates had a comparable number of active accounts on Instagram (58% of candidates) to X (57%). This points to a clear strategic prioritization of X as their primary channel of communication. Liberals and NDP candidates, in contrast, adopted broader multi-platform strategies. While they remained most active on X and Instagram, they significantly increased their activity as the election approached and were also responsible for the modest but notable entry of politicians onto Bluesky. Among Liberal candidates, 73% were active on Instagram, 57% on X, and 22% on Bluesky. In contrast, NDP candidates had a lower participation rate on social media overall with 46% active on Instagram, 21% on X, and 16% on Bluesky. The Bloc Québécois followed a similar reliance on X, though at a smaller scale, with their activity marked by a sharp but short-lived surge in late April that appears to have been event-driven rather than sustained. For Bloc Québécois candidates, 61% were active on Instagram, 49% on X, and 18% on Bluesky. Meanwhile, TikTok and YouTube saw very limited activity across all parties, underscoring how video-centric platforms remain underutilized in Canadian federal politics. Across every party, however, there was a steep decline in posting
immediately after the election, with volumes on X, Instagram, and Bluesky all falling to a fraction of their campaign peaks. This sharp drop-off highlights the campaign-driven nature of political social media activity, with the posting being tactical and time-bound rather than sustained.
We observed changes in engagement trends between the Conservative and Liberal parties throughout the campaign. Throughout 2024, Conservative politicians generally received 2.5 times as much engagement as Liberal politicians (62.0 million likes vs. 24.9 million likes). During the period from February 23–April 28, this gap narrowed to Conservative politicians receiving 1.3 times more engagement (24.8 million likes vs 18.6 million likes). After the election, April 29–May 28, the trend flipped and Liberal politicians received slightly more than Conservative politicians (4.1 million likes to 3.7 million likes). However, more time will be needed to confirm whether the post-election trends hold in the long term. This demonstrates a potential shifting media landscape which aligns with public opinion/polling on partisanship.
Influencers dominated almost half of the online discourse (47% of content posted), followed by news outlets (28%), politicians (18%) and others. Influencers have maintained their spots as crucial political players, shaping public opinion and driving the direction of political conversations. Along with traditional
media outlets such as CBC and CTV, media organizations such as Narcity, Cult MTL, blogTO have risen in popularity online. The latter alternative media organizations represent the rise of "infotainment" or "soft news" sources, which present news alongside social media entertainment. This shift towards soft news is also linked to the [news ban in Canada](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada) which limits the presence of traditional news sources on Meta platforms, giving way for alternative social media actors to fill an information gap online.
### **2.2.2. Thematic analysis of candidates' social media keywords and topic analysis of the Canadian information ecosystem**
In this section, we analyze the content of the social media posts made during the election period. First, we describe the topics and keywords used in the Canadian media ecosystem. We show that election mobilization efforts, the Israel-Palestine conflict, and economic policy issues were the top topics across the Canadian media ecosystem. We also find that both the Liberal and Conservative candidates frequently mentioned "Mark Carney," highlighting the positive anticipation towards the new party leader from Liberal politicians and a significant volume of negative criticism from Conservative politicians. Lastly, we analyze the differences in topics discussed by the gender of the candidates. While male and female politicians receive similar levels of engagement, male politicians are more likely to discuss topics of security, immigration, and political figures, while female politicians discuss topics of social issues, care work, and community.
**Topics during the election and party keywords:** Topic modelling explores the major claims and narratives and their engagement trends during the electoral period. This is a machine learning method that is used to assess all posts, identify clusters of similar content, and then give clustering themes under a label using the most representative words. The clusters are represented on a two-dimensional plane with similar themes placed closer to each other. These themes are subsequently manually explored for further contextualization; between Feb. 23 and May 28, 2025, we analyzed over 1.5 million social media posts.
At large, the top 10 topics ranked by the highest count of posts across the Canadian media ecosystem (see figure 9) are as follows: 1) election awareness and encouraging Canadians to go vote; 2) the Israel-Palestine conflict; 3) support for election campaign volunteers; 4) addressing taxes, debt, and affordability; 5) partisan critique of the Liberal and Conservative parties; 6) Pierre Poilievre; 7) tariffs and trade deals, 8) GDP and economic growth; 9) Trudeau; and 10) Canada-U.S. relations. Other prominent policy issues include health care, transit, housing and affordability, energy and pipelines, and the carbon tax. Although concerns around Chinese interference were prevalent for a one week period (around March 31) after Liberal candidate Paul Chiang was criticized for suggesting that China should claim the political bounty on Joe Tay (see Section 3.1.2), foreign interference and influence were not amongst the most discussed topics within this period at large.
Influencers focused on party politics and the Canada-U.S. relationship. Influencers often discussed the merit of the Liberal and Conservative Parties and their two party leaders, focusing more on partisan critiques and controversial leadership ties, rather than policy proposals and platforms. Influencers also discussed the deterioration of relations between Canada and the U.S., focusing on the Trump administration. When it came to the topic of the Israel-Palestine conflict, which was isolated from the rest of the top ranking topics, influencers and activists largely dominated this discussion. News outlets provided updates on the conflict, relating it to other international affairs such as the war in Ukraine. News outlets also focused largely on key policy issues such as tariffs, housing and affordability, climate, economic growth, and health care. Furthermore, news outlets brought awareness to the election cycle and the platforms of different partisan campaigns.
As for politicians, economic interests remained top of mind. This included discussion on GDP, tax cuts, debt, addressing the oil and gas sector and investment into energy initiatives, the U.S.-Canada trade war, and tariffs. Partisan candidates also engaged in cross-party critique. We investigated further into politician posts by party using keyword analysis, which visualized the most frequently used words from social media posts made by candidates from the Liberal and Conservative parties. Through the keyword analysis, we saw that both the Liberal and Conservative party candidates emphasized Carney in their content. Liberal candidates showed support and anticipation towards him, while Conservative candidates directed criticism towards him. Notably, Conservative candidates spent more time talking about Carney than on supporting their own party leader, Poilievre.
#### **2.2.3. Gender and political messaging in Canada**
[Figure 10](#page-21-0) highlights significant gender disparities4 among representation for candidates of major Canadian political parties. Notably, the NDP and Green Party show higher female representation (52% and 40%, respectively), while parties like the Conservative Party and PPC have a notably higher proportion of male candidates (75% and 79%, respectively). This suggests varying degrees of gender inclusivity or emphasis across parties.
**Figure 9. Main topics discussed in the Canadian Information Ecosystem during the eleciton period (Feb 23 to May 28 2025)**
These two bar charts offer a clear and compelling comparison of how male and female political candidates in Canada engage with different topics on social media. The graphs display the top 10 topics with the highest proportional representation per gender, normalized by the number of candidates, to avoid bias from unequal gender distribution.
The topics most associated with male candidates, as shown in [Figure 11,](#page-21-1) tend to focus on security, immigration, political figures, law enforcement, and crises. Notable themes include: immigration and deportation, law enforcement and crime, COVID-19, drug crises, and foreign policy issues related to Donald Trump and Chinese foreign interference. This suggests that male candidates are comparatively more likely to emphasize topics that are conventionally characterized as masculine, such as national security, crime, and public safety. In contrast, the top topics among female candidates tend to center around social issues, care work, community, and cultural themes. Key topics include: health care and caregiving, gender and LGBTQ+ rights, mentions of family, youth and volunteering, cultural and artistic awareness, and environmental issues. These themes point to a communication style that emphasises conventionally feminine issues and characteristics, such as empathy, social responsibility, and community connection.
**Figure 10. Differences in gender distribution of political candidates by political party.**
**Figure 11. Top 10 topics discussed on social media by candidate gender.**
## **Top 10 topics by candidate gender (normalized per candidate)**
### **Preferred topics for male candidates (by %)**
### **Preferred topics for female candidates (by %)**
Despite some differences in topic preferences and candidate representation, engagement metrics reveal remarkable parity. Female candidates, while posting less frequently, received engagement levels nearly identical to their male counterparts. Average likes per post are nearly the same (66 likes per post for male candidates, 63 for female candidates).
While the differences in topic engagement by gender are relatively minimal (falling between 40-60% in [Figure 11](#page-52-0)), this pattern illustrates that there may still be a moderate trend of gendered division in political discourse, whereas male candidates are more often linked to "hard" policy areas and external authority roles, women are associated with "soft" topics that focus on caregiving and community. Gendered division in political communication can reinforce stereotypes of the role of political candidates, where male candidates are perceived as more competent in authoritative and handling leadership positions that address concrete policies. This can also constrain the perception of female politicians and their public representation. Regardless, comparative engagement between female and male candidates indicates equal audience appreciation and interest in both demographics of candidates as viable elected officials.
### **2.2.4. Influencers linking media, politics, and the public**
To explore patterns and overlap in political messaging and information diffusion across party lines, we conducted a semantic network analysis using social media posts (on X, YouTube, TikTok, Instagram, Bluesky and Telegram), as seen on page 18 (figure 12). Each node in the network represents a single post shared within a 12-hour window. Posts were connected if they reached a semantic similarity score of 0.95 or higher (indicating very similar content and structure), based on sentence embedding comparisons using a transformer-based model.
Unlike traditional retweet networks, this approach captures content-level alignment — even when identical messages are shared without formal reposts. Reposts were retained, as they were interpreted as both amplification and endorsement of the message. Posts with higher PageRank are more semantically similar to a broader set of other posts, and thus more central to the overall flow of information. The resulting network exhibits distinct content clusters that largely align with political party boundaries, reflecting ideologically coherent messaging and internal coordination. However, a small but crucial subset of nodes — Canadian influencers (yellow) — emerges as central bridges in the network. The size of each node reflects its Page-Rank score5 (the quality/importance of a page as determined by counting the number of links that lead to it), indicating its relative centrality and influence within the network. The names listed in figure 12 illustrate the prominent Canadian influencers central to the network.
Unlike formally partisan actors, these influencers connect content across otherwise disconnected communities, including media outlets, political parties, and foreign-affiliated accounts. Their central position in the network suggests that they serve a mediating role, enabling the diffusion of ideas beyond partisan silos and throughout the media ecosystem.
Interestingly, while most party-based clusters remain ideologically self-contained, both influencers and foreign accounts (black nodes) appear to cut vertically across the network structure. This visual pattern breaks from the classic echo chamber configuration and provides strong evidence of how certain actors facilitate the flow of content between otherwise isolated political groups. It highlights not just who is talking, but who is enabling content to move across ideological divides.
### **2.3. Avatar-based social media monitoring**
Our "boots on the ground" social media monitoring team used 39 social media avatars to monitor the online media environment daily during the election period. These monitoring efforts revealed a significant divergence in the electoral discussions on social versus traditional media platforms. As described in detail throughout this section, online content prioritized attention-grabbing narratives over substantive debate about party platforms or policy proposals. Public focus centered on the cost of living, crime, Canada-U.S. relations, and, to a lesser extent, identity-based grievances.
Key findings from this section indicate that what Canadians encountered in their feeds during the 2025 campaign was fast-moving, polarized, and shaped more by attention dynamics than by platformed policy debate. Leaders' content, especially from Carney and Poilievre, was ubiquitous, while Singh's visibility clustered in younger and left-leaning spaces. Right-leaning and centrist feeds were saturated with meme-heavy frames, media criticism (notably of the CBC), and, as election day neared, escalating fraud allegations, while left-leaning feeds amplified strategic-voting appeals and issue activism around Israel-Gaza and climate.
Election day itself saw a sharp spike in narratives about fraud, U.S./Trump influence, crowd-size signaling from rallies, the Vancouver Lapu Lapu festival tragedy reframed through partisan lenses, and high-volume voter-information pushes. Post-election, right-leaning streams focused more on fraud and foreign interference while left-leaning streams pivoted to celebration, coalition-building, and policy critique, alongside concern over renewed talk of Western separation.
**Figure 12. Semantic diffusion of political messages.**
Platform dynamics were distinct: X functioned as the most political and disinformation-dense venue; TikTok exhibited strong algorithmic siloing; Facebook groups reflected echo chambers and economic-nationalist themes among older users; Instagram was used heavily for community mobilization.
Finally, interaction patterns were marked by homophily (ideological clustering) and "trench" replies (where users from opposing sides directly confronted one another) rather than deliberation, with identity cues reinforcing in-group cohesion.
Note that this analysis is reflective of the experience of the research team and their use of their avatars. See the [Limitations](#page-61-1) section for a discussion of the generalizability of these findings.
### **2.3.1. Dominant narratives**
In the lead up to the election (the election period), our avatars identified six key patterns:
1. **Content directly from party leaders** appeared frequently across all avatar accounts, with Carney and Poilievre the most heavily featured. Singh also appeared in several feeds, but his visibility was more limited and largely confined to left-leaning and younger online spaces. Overall, Liberal MP candidates were frequently encountered (independent of avatar partisanship), largely due to the high visibility of current Liberal ministers and the PM himself, followed by Conservative Party candi-
dates. Content from NDP candidates was only encountered by avatars with a clear affinity with the party. These trends emerged around the time candidate lists were being finalized.6
- 2. **Poilievre's platform**, emphasizing tax cuts, anti-'wokism', and crime policy, dominated right-leaning and centrist feeds, often in clusters of multiple posts. Posts often framed Poilievre as restoring Canada and reversing what was described as a Liberal-induced economic decline. Notably, this content appeared not only as formal campaign posts, but also as influencer content, commentaries and memes.
- 3. **Media criticism**, especially towards the CBC, was featured on both centrist and right-leaning feeds. Some posts also targeted CTV and The Globe and Mail for biased coverage and calling the election result too early, but the CBC remained the primary focus. Notably, the nature of engagement on this topic is polarized: Some posts defended the CBC as a critical source of public journalism under attack, while others portrayed it as a biased and partisan source of information for the Liberals.
- 4. **Strategic voting** (e.g., vote NDP/Green to block the Conservative Party) also gained popularity closer to the election date, especially among left-leaning avatars. These posts encouraged audiences to vote NDP or Green in certain ridings to block Conservative wins, often using electoral maps, polling data, or community-specific calls to action. While not always formally tied to party operations, these narratives circulated organically and rapidly, pointing to a grassroots coordination effort among progressive communities online.
- 5. **Israel-Gaza** was pervasive on left-leaning and younger avatar feeds, particularly on TikTok and Instagram. There was a consistent level of visibility of pro-Palestine activism, including posts calling for ceasefires, criticizing Canadian complicity, and linking the conflict to broader anti-imperialist rhetoric. While some avatars did encounter pro-Israel content or attempts at more neutral overviews of party positions, these were less common.7
- 6. **Climate/environment content** was also dominant on left-leaning and younger avatar accounts. Discussion included posts calling for more actions on the climate crisis, promoting electric vehicles, and criticising the oil and gas industry were also highly visible on left-leaning feeds. This discussion was mostly driven by the general public and less so by influential accounts.
On election day, we observed six key themes:
- E **Election fraud** allegations were, by far, the topic most widely discussed across platforms (including the pen/pencil debate, early media calls, and the longest ballot as subsequently discussed in Section 4.1.2 on misinformation). While claims were almost exclusively generated by rightwing influencers, they were also encountered by centrist/ apolitical avatars.
- E **U.S. and Trump influence** was also a prevalent topic of conversation, especially when President Trump posted a statement about the 51st state on Truth Social on election day. Some posts featured a mockery of the statement, while others showed outrage against Trump's influence.
- E **Election day rallies in Carleton and Edmonton** triggered significant discussion on crowd size, and many posts with footage were shared documenting the experience. Many posts used footage, applause lines or emotional imagery such as family moments to signal the momentum.
- E **The Lapu Lapu festival tragedy**, where a vehicle rammed into a crowded street in Vancouver, also generated considerable election day discussion. Most posts framed the incident through a partisan lens, with some praising Carney for cancelling a campaign rally and others criticising Poilievre for politicizing the tragedy.
- E **Election information and voter mobilization** posts were frequently encountered on election day. Posts on polling locations, rights and the civic process were shared mostly by influencers and official accounts (like Elections Canada) focused on. Meanwhile, posts encouraging mobilization and strategic voting were most common on left-leaning and youth avatar feeds.
- E More broadly, **we observed significant variation in the tone of discussion across avatars on election day**. Comparing left, right and centrist leaning avatars, right-leaning feeds were highly emotional and meme-driven, sometimes featuring conspiratorial tones. Left-wing feeds, in contrast, tended to be more mobilization-focused (e.g., #hotgirlsvoteliberal). For the centrist and apolitical social media feed, we observed a high number of informational posts, mixed with narratives of election fraud.
In the week following the election, we noticed an increase in election-fraud narratives (see Section [4.1.2\)](#page-42-1). Right-leaning feeds remained focused on fraud claims, media mistrust, and foreign interference, while left-leaning profiles featured more celebration, coalition-building messages, and policy critiques of Conservative positions. Calls for Western separation, especially from Alberta and Saskatchewan, also gained traction.
#### **2.3.2. Comparing platform discussions**
Across all platforms, qualitative observations from the Avatar team found that X stood out as the most political and disinformation-heavy space, consistently surfacing election-related content for right but also centrist and apolitical avatars. X was a major source of election fraud claims, foreign interference accusations, and AI-generated visuals, often delivered with a conspiratorial tone. TikTok, while culturally meme-driven, showed strong algorithmic siloing, pushing partisan content based on age and political leanings. Facebook had strong echo chamber effects (for Facebook groups, the content is mostly curated by power users who have a specific view of things), with recycled narratives from X and TikTok and strong themes around "Buy Canadian" and economic nationalism, particularly among older users. Instagram showed the least organic political content (that is, user-generated political posts shared without paid promotion or influencer coordination), dominated by lifestyle and ads, though some far-right content surfaced late in the campaign.
#### **2.3.3. Comparing dominant actors**
Narratives expressed by political parties and influencers were clearly divided along partisan lines. Political parties pushed clear and consistent narratives to the Canadian public during the election period. Comparing the two major parties, Conservative and Liberal, the Conservative party tended to promote "Canada is broken" attributing this perceived decline to the policies and legacy of the Liberal government under Trudeau, with Carney positioned as a symbolic continuation of that era. Their communications emphasized a return to a more traditional Canada, characterized by home ownership, public safety, and value-neutral governance. Meanwhile, the Liberal party projected itself as forward-looking, framing the party as the builder of a future-ready Canada. Carney's messaging focused on job creation, infrastructure development, and housing, situating the party as a stabilizing force amid shifting geopolitical dynamics, particularly with the United States. This framing positioned the Liberals as the party of economic resilience and adaptation in the face of global uncertainty. While other parties, like the PPC and NDP, had limited agenda-setting power, they still actively appealed to their supporters. The PPC focused on framing the dialogue around conspiratorial narratives, criticising both the Liberal and Conservative parties for not representing Canadians. The NDP focused on humanitarian issues, health care and the corporate dominance in Canada (e.g., criticism toward Loblaws).
By contrast, influencers and influencer organizations tended to jump between trending issues, switching from debates over foreign interference to national media to election integrity,
while maintaining clear partisan leanings. For example, after the leadership debate, right-wing voices would pivot across those topics yet invariably frame the CBC as a mouthpiece for "liberal elites," whereas their left-leaning counterparts hopped through the same issues but consistently praised the CBC as an essential, neutral arbiter. This approach appeared well-aligned with platform algorithms that rewarded emotionally engaging or polarizing content. Among politically disinterested users, content was often shaped by influencer styles of delivery (fastpaced, emotionally framed posts that are often sensationalist) rather than direct party communication.
#### **2.3.4. Engagement overview**
We saw relatively high levels of homophily, particularly on X, and to a lesser extent on Meta platforms. Users primarily engaged with content aligned with their political views, and even seemingly neutral narratives, such as Liberal messaging about resilience against U.S. pressure, were often interpreted through a partisan lens. Interpartisan engagement did occur, but it was frequently adversarial or dismissive rather than deliberative.
Some users employed visual markers, such as national or identity-based flags and emojis, to signal political or ideological alignment. A notable example was the emergence of the "" (apple) subcommunity among Conservative supporters on X. This simple symbol quickly became a badge of in-group identity — members amplified each other's content, reinforcing the pattern of clustered engagement without broader outreach.
Comment sections under polarizing posts often revealed "trench" dynamics. While this reflects a degree of cross-ideological interaction, it has rarely translated into meaningful dialogue.
## **2.4. Incident response**
Throughout the duration of the electoral period, the Incident Response Team detected and responded to a wide variety of potential, minor and moderate severity information incidents (defined here as any coordinated or deliberate effort to manipulate public opinion, distort facts, or sow confusion) during the writ period of March 23 to April 28. Potential incidents (which can be considered as *threats*) were investigated for evidence of disruption; if disruption was significant enough, it was declared and triaged as a minor, moderate or major incident. In the process of monitoring the ecosystem, we identified threats in various forms, including suspicious, contentious and malicious narratives, accounts, platform activities, and online behaviours.
In total, we identified 35 incidents (potential and actual). 21 remained as potential incidents, i.e., information gathered indicated insufficient evidence to merit escalation to a minor or moderate level based on our [incident response protocol](https://www.cdmrn.ca/incident-protocol). Threats ranged from disinformation campaigns related to party platforms or candidate ties to foreign entities, to the use of online platforms to organize campaigns geared towards domestic unrest (firearms rights groups, 51st state groups). We classified 14 of these incidents as actual, with 12 minor (Level 1) and 2 moderate (Level 2). Minor information incidents largely focused on alleged cases of foreign influence campaigns by China and the U.S. as well as AI-generated or manipulated content. Moderate incidents, in contrast, focused on fake news content linked to cryptocurrency scams and election fraud. Both were escalated based on their potential to sway voters and influence trust in the integrity of the election. The first moderate incident (see [Sec](#page-39-1)[tion 4.1.1](#page-39-1)), "[AI-Generated Fake News](https://www.cdmrn.ca/ai-generated-fake-news)," involved a surge in fake news articles linked to cryptocurrency scams on Facebook. This incident demonstrated a potential to mislead Canadians into faulty investment schemes and influence public perceptions of political candidates, media organizations, and, more broadly, the electoral process. The second moderate incident (see [Section](#page-42-1) [4.1.2](#page-42-1)), "[Surge in claims actively undermining election integ](https://www.cdmrn.ca/weekly-updates/2025-05-01-en)[rity](https://www.cdmrn.ca/weekly-updates/2025-05-01-en)" (reported in detail in our last weekly update), followed the detection and analysis of a series of election fraud-related narratives that collectively threatened trust and the stability of the final days leading up to and actions on election day. Fortunately, we did not identify any major incidents (Level 3). For details on the full scope and nature of incidents identified during the election, see [Section 4](#page-37-1) on Information Incidents.
# **2.5. Story Chaser and public tipline**
To monitor and analyze information incidents during the 2025 Canadian federal election, we used two complementary data collection solutions: the Story Chaser system, an internal tool used for structured tracking of potential information incidents identified on social media, and a public-facing digital threat tipline, which collected over 238 submissions from Canadians and the general public flagging concerning content they considered as threatening to the election in their social media feeds.
Over 300 entries tied to 35 flagged information incidents were logged in the internal Story Chaser system during the election period. Consistent with other data collection methods, dominant narratives focused on election fraud, foreign interference, and elite political figures, particularly Carney. X emerged as the primary platform for real-time disinformation propagation, surpassing Facebook, TikTok, and Instagram in monitored activity. Threats were frequently associated with electoral integrity, partisan framing, and conspiracies involving ballot tampering and foreign financing. Frequently used metatags, such as *election\_fraud*, *can\_media* (Canadian Media), *foreign\_finance*, and *gen\_ai* (Generative AI)*,* highlighted both persistent concerns and emerging risks. Overall, the data reflected a polarized and distrustful digital landscape in which media skepticism and international influence remain key concerns.
Public submissions through our digital threat tipline revealed election and voting integrity as the top concern, followed by reports of manipulation, foreign influence, and election actor behavior. A submission spike between March 22 and April 1 revealed heightened civic alertness, with Facebook appearing in over 60% of reports, far surpassing other platforms like Instagram, YouTube, TikTok, and news sites. Submissions ranged from clear, fact-based reports to subjective expressions of distrust in political figures and media, blending personal perception with threat identification. This convergence of disinformation awareness and political interpretation reveals both the public's growing vigilance and its susceptibility to characterizing bias or political disagreement as manipulation. Common language used in reports reflected not just concerns about falsehoods, but about their social circulation and impact. These findings highlight a dual challenge: maintaining civic vigilance while strengthening public understanding of what constitutes verifiable manipulation in an evolving threat landscape.
# **3. Election integrity**
Safeguarding the integrity of Canada's democratic process requires careful attention to a wide array of threats, both real and perceived. In this section, we evaluate election integrity through multiple lenses, drawing on survey data, digital trace evidence, and incident analysis. Our framework considers four key dimensions: the role of foreign influence and interference, the spread of misinformation and disinformation, the presence of bot-driven or otherwise inauthentic activity, and the disruptive potential of generative artificial intelligence. This section contains examples of each of the four, with further incident-specific details in the following subsection.
Key findings from this section include that Canadians perceived foreign interference as a growing threat, with concern about U.S. meddling nearly tripling since 2021 (largely driven by Trump's annexation rhetoric, tariff disputes, and the deep entanglement of U.S. politics in Canadian discourse). China-related narratives were a persistent feature of the campaign, centering on alleged ties between Canadian politicians and Chinese state-linked actors and amplified by controversies involving Joe Tay, Paul Chiang, and Carney; while their overall impact was modest, these episodes highlighted vulnerabilities within diaspora communities and the risks of transnational repression. Russia, meanwhile, played a visible role in the information environment by promoting anti-woke narratives, pushing disinformation about Carney, and amplifying issues such as Canada's support for Ukraine. Although the reach of these efforts was limited, they reinforced partisan divides. Concerns about Indian interference also surfaced, largely echoing earlier allegations tied to the 2022 Conservative leadership race, but activity during the 2025 campaign was minimal compared to that of the U.S., China, or Russia. Beyond foreign interference, misinformation was prevalent but had limited measurable electoral impact, while bot-driven activity on X amplified conspiratorial narratives. Generative AI added a new layer of risk through deepfakes, fabricated articles, and manipulated visuals that fueled doubts about the reliability of political information. Taken together, these findings underscore that while the integrity of the election was not compromised, the threat environment has become more complex, with foreign interference fears, platform manipulation, and technological disruption shaping both citizen perceptions and the broader information ecosystem.
# **3.1. Foreign influence**
Throughout the electoral period, we sought to identify instances of foreign interference and influence activity from countries including the United States, China, India, and Russia. In this context, *foreign interference* relates to the covert technical and clandestine targeting of electoral processes and systems to fulfill a foreign actor's interest at the expense of another nation's interest, while *foreign influence* relates to overt efforts to directly or indirectly shape election processes or outcomes through mechanisms such as public opinion, funding of diaspora institutions, and diplomatic pressure or lobbying, to advocate for a foreign actor's interest. Overall, we found modest interference and influence efforts directed at the Canadian election. We observe that Canadians were very concerned with foreign interference and there was significant speculation and discussion of foreign interference during the election. A vigilant population is critical to limiting the impact of any interference or influence attempts, and this vigilance likely contributed to the low volume observed.
Among the Canadian public, perceived threats of foreign interference were significantly higher during the 2025 election than in 2021 (see [Figure 13 o](#page-28-0)n page 23 showing this comparison). This upward trend is not surprising given the attention that foreign interference has received in the months and years preceding the election, with the release of the [Foreign Interference Com](https://foreigninterferencecommission.ca/reports/final-report)[mission's final report](https://foreigninterferencecommission.ca/reports/final-report) in January 2025 and revelations about foreign interference in the most recent [Liberal](https://www.cbc.ca/news/politics/freeland-wechat-malicious-activity-1.7454067) and [Conserv](https://globalnews.ca/news/11097709/poilievre-leadership-india-meddling-allegations/)[ative](https://globalnews.ca/news/11097709/poilievre-leadership-india-meddling-allegations/) leadership races. However, the most significant trend
**Figure 13. Comparison of public perceptions that countries engaged in foreign interference during the 2021 and 2025 elections.**
concerns Canadians' perceptions that the United States would try to meddle in the election, particularly in light of Trump's escalating rhetoric and threats of annexing Canada through "economic force." As a result, as shown in [Figure 13,](#page-28-0) the percentage of Canadians who believe that the U.S. interfered in the election jumped from 17% to almost 50% between the 2021 (September to October) and 2025 (May) post-election surveys. Smaller but significant increases were also observed for China (from 21 to 28%) and Russia (from 12 to 20%).
During the 2025 election, perceptions of foreign interference varied markedly across partisan groups but were quite stable during the campaign (see [Figure 14\)](#page-55-1). Liberal, NDP, and Bloc Québécois supporters were most concerned about the United States, consistently reporting higher perceptions of U.S. interference than the Conservatives. Liberals also expressed more concern about Russia than Conservatives. Conservatives, by contrast, identified China as the dominant threat, with consistently higher levels of concern than any other group, with lower evaluations of the risks posed by the U.S. and Russia. NDP supporters showed a mixed profile, aligning with Liberals in their elevated concern about the U.S. but also registering moderate and relatively balanced concerns about China and Russia. Bloc Québécois supporters presented a more even distribution of threat perceptions, with China, Russia, and the U.S. all viewed as significant sources of interference, while India was consistently regarded as a lesser concern. Overall, the data
reveals a modest partisan divide: Liberals and the NDP are most concerned about U.S. interference, Conservatives (and PPC supporters) are most concerned about China. Also notable is that there were few major shifts across the campaign, with relatively stable perceptions of interference for all party-country pairings.
Using our digital trace data, we also assessed the prevalence of foreign interference discussions in our seedlist posts. Specifically, we used an automated method involving keyword search and large language model (LLM) validation to identify any posts discussing foreign interference as well as the potential entities who were claimed to be responsible. Of the 1.2M relevant seedlist posts between Feb. 23 and May 28 2025, 3,524 were identified as discussing foreign interference. Examples of these discussions are shown in [Screenshots 1](#page-29-0).
[Figure 15](#page-55-2) shows the distributions of these posts by source and platform. There are marked differences between the perceptions from our survey and observed online discourse. Notably, discussions of Chinese FI were by far the most common with a large number of posts on X and YouTube. While perceptions of American interference were highest on our survey, they were only the third most discussed source among seedlist posts, behind India and China. Similarly, discussions of Russian interference were notably much smaller compared to Chinese, Indian, and American FI even though perceptions on our survey had Russia much higher. Note here, that "general" interference represents posts that claimed foreign interference without specifying an
**Figure 14. Public perceptions that countries attempted to influence Canadian politics and elections by partisanship.**
exact source; for example by referring generically to "hostile" or "foreign" countries. Our analysis also identified the World Economic Forum as a source of Foreign Interference being discussed. Other more minor countries included Israel (just 10 posts) and Pakistan (just 6 posts).
We observed variance in attention paid to foreign interference. For China, the three most prominent peaks correspond to March 25 (news of Carney's Brookfield loans to the Chinese Central
Bank broke), March 31 (Tay/Chiang incident) and April 10 (the Globe and Mail story of Carney having met with a Chinese front group for the United Front Works Department). For India, a peak was observed on March 25, when Canadian Security Intelligence Service (CSIS) published a report linking India to the Poilievre leadership race. For the United States, a small peak on March 23 is the day a Breitbart interview with Alberta Premier Danielle Smith resurfaced and went viral. In the interview, the premier discussed how the U.S.-Canada trade war was boosting polling
### **Screenshots 1:**
**Influential Canadians posting about foreign interference in the lead up to and during the 2025 Canadian federal election.**
**Figure 15. Top 7 sources of foreign interference as discussed in posts from our seedlist.**
numbers for the Liberals, which led some outlets to claim she was encouraging the U.S. to interfere.
We examine each of the four major countries below.
### **3.1.1. United States**
In the months leading up to the Canadian election, a series of actions by the United States deepened anxieties about foreign influence. President Trump repeatedly floated the idea of annexing Canada, first suggesting it become the ["51st state" in](https://ici.radio-canada.ca/rci/en/news/2129899/whats-really-behind-donald-trumps-tariff-threats-and-51st-state-posts-about-canada) [late 2024](https://ici.radio-canada.ca/rci/en/news/2129899/whats-really-behind-donald-trumps-tariff-threats-and-51st-state-posts-about-canada), then escalating in January 2025 by declaring he would use ["economic force" to achieve this aim](https://ici.radio-canada.ca/rci/en/news/2131198/trump-says-he-would-use-economic-force-to-join-canada-with-u-s). His inauguration speech revived [American expansionism rhetoric](https://nationalpost.com/news/donald-trump-referred-to-manifest-destiny-in-his-inaugural-address-heres-what-it-means). At the same time, the White House pursued an aggressive tariff agenda in announcing sweeping 25% tariffs on Canadian goods and later targeting steel and aluminum, which amplified economic uncertainty. U.S.-based groups like the [America Fund](https://www.cbc.ca/radio/asithappens/bowden-51st-state-billboard-1.7468264) and the Atlas Society also ran billboard and social media campaigns in Canada, framing statehood as desirable while amplifying grievances over federal policies. Prominent U.S. officials, including Peter Navarro and White House Press Secretary Karoline Leavitt, made inflammatory statements about [redrawing the Canadian](https://www.telegraph.co.uk/us/politics/2025/02/27/white-house-canadian-border-trump-trudeau/?msockid=14d5e314491e616b2e02f6f448316021) [border](https://www.telegraph.co.uk/us/politics/2025/02/27/white-house-canadian-border-trump-trudeau/?msockid=14d5e314491e616b2e02f6f448316021) or questioning [Quebec's viability](https://www.lapresse.ca/international/etats-unis/2025-03-12/annexion-du-canada/la-maison-blanche-utilise-les-impots-du-quebec-pour-se-justifier.php). Collectively, these actions blurred the line between legitimate bilateral disputes and overt interference, raising fears that American political rhetoric and digital influence campaigns were shaping Canadian public debate.
Despite the pressure, there was significant support across Canada for resisting these incursions. Federal leaders presented a united front, with both government and opposition figures condemning American rhetoric as a direct threat to Canadian sovereignty. Civil society organizations and provincial leaders likewise mobilized, rejecting suggestions of annexation and emphasizing national unity. Canadian trade retaliations to tariffs signaled a willingness to defend economic interests, while grassroots movements highlighted pride in Canada's independence and democratic institutions. This broad support was reflected in the attitudes of Canadians and in the seriousness of the response.
As highlighted in [Figure 16,](#page-31-1) very few Canadians (5%) believe that Canada should become the 51st state of the United States. However, Trump's comments on the matter are generally perceived as a serious threat, with 59% perceiving them as such. Few (12%) believe that Trump has Canadians' best interests at heart when discussing the potential annexation of Canada. Instead, most Canadians believe that access to our natural resources (87%), his ego (85%), and geopolitical and military reasons (77%) constitute his primary motivations.
While condemnation and criticism were widespread, they were not universal. We assess a marked surge in activity among both previously dormant and newly created social media pages and groups that call for the annexation of Canada to the U.S.
**Figure 16. Canadian public opinion regarding the 51st state.**
Several Canadian influencers publicly posted support for increased integration with the United States, with [one even ap](https://www.westernstandard.news/news/watch-alberta-as-51st-state-advocate-appears-on-fox-news/62839)[pearing on Fox News](https://www.westernstandard.news/news/watch-alberta-as-51st-state-advocate-appears-on-fox-news/62839). Some [polls and reports show](https://www.thefp.com/p/the-canadians-who-want-to-be-american) as many as 10% of Canadians supporting annexation.
During the election, we closely followed advocacy efforts for 51st statehood, leading us to investigate [51st state Facebook groups](#page-49-1) and online advocacy for Alberta's separation or annexation. While these movements exhibit elements of authentic grassroots support, analysis revealed that they intensified their activities during the federal election period, employing techniques such as astroturfing to create the illusion of broader backing and promoting narratives of election fraud within online communities. Notably, leaders of these movements maintain documented ties to the Trump administration and, in some cases, their affiliated social media pages are administered by U.S. citizens, indicating cross-border cooperation. Although these initiatives appear not to have significantly disrupted the broader Canadian media ecosystem, they are concerning in both their online and offline presence.
### **3.1.2. China**
Concerns about Chinese interference in Canadian politics were already prominent leading into the election, largely due to sustained media coverage and the [Public Inquiry](https://publications.gc.ca/collections/collection_2025/bcp-pco/CP32-169-2-2025-eng-3.pdf) into Foreign Interference launched in 2023. The inquiry, initiated in response to intelligence leaks and mounting political pressure, investigated allegations that Chinese state-linked actors sought to influence the 2019 and 2021 federal elections. Its interim findings, released in early 2024, drew widespread public attention by highlighting attempts to covertly influence political candidates and to monitor or pressure members of Chinese-Canadian communities. This set the stage for heightened public sensitivity around China-related narratives during the 2025 campaign.
There was significant concern surrounding China's alleged interference in the Canadian election. While online discourse on the issue remained limited as compared to other topical issues, such as tariffs and affordability, engagement on the matter increased during allegations of political candidates' connections to the Chinese government in early April. Public concern and attention over Chinese influence grew during this time. Several political figures, including Tay, Chiang, Peter Yuen, Wilson Miao, Chrystia Freeland, and Carney were linked to China-related controversies, with [one documented instance](https://www.cbc.ca/news/politics/freeland-wechat-malicious-activity-1.7454067) of WeChat-based interference in January-February.
Alerts most commonly centred on allegations of financial, or clandestine and organized connections between political candidates or parties and Chinese state-linked entities. Notable examples include an investigation into pro-Liberal Party sentiments being disproportionately amplified influence on WeChat, which was flagged by the Security Intelligence Threats to Elections (SITE) Task Force (a whole-of-government working group that coordinates the Government of Canada's collection and analysis **Screenshots 2: Influential Canadians posting about the annexation of Canada by the United States.**
efforts concerning threats to federal election processes), the [Repression Operation](https://www.cbc.ca/news/politics/joe-tay-china-campaign-1.7514934) against candidate Tay (also flagged by the Task Force), as well as scrutiny of Carney's economic ties to China.
Notably, discussion over potential Chinese interference peaked in late March and early April, following a major controversy involving the then Liberal MP Chiang and Conservative candidate Tay. In a Chinese language interview in January, Chiang said that Tay, a political dissident known for his pro-Hong Kong independence activism (and with a [HK\\$1 million bounty](https://apnews.com/article/hong-kong-activists-arrest-bounties-6d8e476341c93969a8d32ba43a61151f)), should be handed over to the ["Chinese consulate](https://www.cbc.ca/news/politics/liberal-paul-chiang-bounty-joe-tay-conservative-1.7496751)," a remark that raised serious concerns about transnational repressions. The remark gained wide attention when it was flagged by the RCMP and reported by major news outlets such as the CBC on March 28. Discussion over the incident was widely circulated on social media, many feared the involvement of the Chinese government in Canadian politics and a pattern of interference with Canadian political contests. Online criticism also expanded to target Carney and the Liberal party, with many accusing them of failing to respond efficiently. Chiang's delayed resignation also further contributed to the accusation of Carney having potential ties with China (see [Section 4.2.5\)](#page-51-1).
We also investigated potential incidents of transnational repression targeting Canadians within the Chinese diaspora community, who are particularly likely to use Chinese-language platforms such as RedNote and WeChat.
Given the prominence of Chinese interference discussion, we used automated analysis using a similar methodology to [Figure](#page-30-1) [15](#page-30-1) (Top 7 sources of foreign interference) to group and quantify the main narratives within our seedlist posts. Specifically, posts were labelled as those discussing: 1) ties between China and Carney, 2) controversies surrounding Tay and Chiang, 3) Chinese Wechat interference flagged by the SITE force, and 4) any other interference involving China. The classification was
multi-label since posts could contain more than one narrative.
[Figure 17](#page-33-0) shows the results, primarily indicating that the most discussed claim was about ties between Carney and China. Of the 2,230 posts from influential voices that discussed Chinese interference, 65% contained discussion of ties between Carney and China, 25% contained discussion of Tay/Chiang, 3.5% contained discussion of the WeChat incident, and 16% contained other discussion of Chinese interference. As with general foreign interference, this discussion and engagement was strongly associated with X. In terms of overlap, 187 posts contained at least two of the three main claims. 64% had overlap between the Carney/China and Tay/Chiang narratives, 27% overlapped between the Carney/China and WeChat narratives, and only 9% had overlap between all three. There was no overlap solely between the Tay/Chiang and WeChat narratives.
We assessed several incidents related to China and determined minor or negligible impact. We note, however, that the Chiang and Tay incident highlights the continuing vulnerability of the Chinese diaspora community here in Canada to Chinese influence.
### **3.1.3. India**
Concerns about potential Indian interference in Canadian politics surfaced early in the 2025 federal election campaign, building on reports that India had attempted to influence the 2022 Conservative leadership race. At that time, the CSIS [alleged](https://www.theglobeandmail.com/politics/article-india-alleged-foreign-interference-pierre-poilievre-conservative/) that Indian state-linked actors raised funds and organized diaspora support to help Poilievre secure leadership, though no evidence suggested that Poilievre or Conservative Party members were aware of these efforts. These revelations set the stage for heightened vigilance in 2025, though ultimately Indian activity was perceived as more limited compared to that of the United States or China. Post-election surveys found that 18% of Canadians believed India had interfered, compared to 49% for the U.S. and 29% for China.
**Figure 17. Breakdown of Chinese foreign interference narratives by platform.**
One significant episode involved misleading reports on the personal finances of key candidates. By March 2025, several Indian outlets, such as the [Times of India](https://timesofindia.indiatimes.com/etimes/trending/is-mark-carney-a-billionaire-know-the-net-worth-of-canadas-next-pm/articleshow/118848684.cms) and [Hindustan Times](https://www.hindustantimes.com/world-news/us-news/mark-carneys-net-worth-how-rich-is-canadas-new-pm-101741561270104.html) republished figures on Carney and Poilievre's net worth that originated in Canada from AI-generated, inaccurate articles. These stories, though debunked by Canadian media (e.g., [CBC](https://www.cbc.ca/news/canada/party-leader-networth-misinformation-ai-1.7498417)), circulated widely and became some of the most visible online content regarding candidate finances. Their amplification by Indian media outlets illustrates how external actors can legitimize questionable sources, embedding falsehoods into the public conversation.
Another focal point of Indian-linked influence operations was the longstanding tension surrounding the Khalistan separatist movement. Indian accounts and Canadian proxies circulated narratives accusing the Liberal Party and [NDP](https://www.tiktok.com/@canamnetwork/video/7434240648993180984?q=liberal%20party%20pro-khalistan&t=1749736621248) of pandering to pro-Khalistan constituencies while framing Conservative candidates, particularly Poilievre, as more aligned with pro-India and Hindu diaspora communities. Influencers such as Shree Iyer of PGurus [explicitly urged](https://www.youtube.com/watch?v=aOWvQbpt7ng&themeRefresh=1) Hindu Canadians to back the Conservatives, portraying Carney as sympathetic to pro-Khalistan groups. These efforts reflect India's broader geopolitical concerns with Canada's stance on Sikh separatism and its accusations that Canadian governments have harbored individuals linked to the Khalistan cause.
The impact of these narratives was most visible in the controversy surrounding Don Patel, the former Conservative candidate for Etobicoke North. Patel was [dismissed](https://www.cbc.ca/news/canada/toronto/conservative-party-candidate-dropped-etobicoke-north-1.7500698) by the Conservative Party in early April 2025 after reports linked him to online posts supporting Indian surveillance of Sikh Canadians and calling for the deportation of Pro-Khalistan Canadians. The incident sparked intense debate on social media among the Indian diaspora communities, with several Indian commentators framing the decision as a betrayal of pro-India voices in Canada. The episode highlighted how diaspora identity politics can intersect directly with election candidate selection, leading to polarization within diaspora communities.
India-linked voices attempted to turn [Poilievre's perceived](https://youtu.be/HQiYmkYMFmM?si=x27m0Ku9z9adUp_O) [neutrality](https://youtu.be/HQiYmkYMFmM?si=x27m0Ku9z9adUp_O) on Khalistan into a strength, appealing to pro-India voters disillusioned with the Liberals and NDP. The result was a fragmented information environment in which competing diaspora-focused narratives, pro-Khalistan versus pro-India, sought to mobilize support along identity lines.
Compared to the United States' overt rhetorical threats and China's systemic engagement with diaspora networks, India's influence efforts were narrower and more fragmented. Their primary vectors were AI-driven misinformation and diaspora-targeted identity politics. Nevertheless, these efforts risked amplifying polarization within Canada's Indian diaspora and undermining broader public trust in the integrity of elections.
The CSIS findings from 2022 remain an important reminder that foreign states may attempt to cultivate long-term influence channels within Canadian politics. In 2025, however, the Indian dimension, while concerning, was secondary to larger interference debates. The challenge for Canadian institutions moving forward lies in building resilience against identity-based disinformation while balancing sensitive diplomatic relations with India.
#### **3.1.4. Russia**
With a well-documented history of electoral interference and democratic delegitimization efforts abroad, Russia was identified as a likely actor seeking to influence Canada's federal elections campaign. These concerns were heightened following [revelations](https://www.cdmrn.ca/russian-funding-canadian-influencers) about Tenet Media, which was created by two Canadians and covertly funded by the Russian state media network, RT. Investigations showed that Tenet Media received nearly USD 10 million to promote Kremlin-aligned, divisive and polarizing narratives through Canadian and American influencers. While it remains unclear how this exposure, triggered by legal indictments and media coverage, impacted Russia's interference strategy during the electoral campaign six months later, the incident remains an important consideration when analyzing Russian actions, and Canadians' perceptions of Russian interference, during the election period. During the election, we assessed several incidents related to Russia and determined minor or negligible impact.
Social media platforms (notably X and Telegram) of the Russian-sponsored media, RT and Sputnik, featured significantly more coverage and reposts of the Canadian federal election than their official websites. Carney and Singh were frequently targets of accusatory or mocking rhetoric with Carney especially in relation to the "head of the 51st state" narrative, while Poilievre and Maxime Bernier (both framed as [anti-woke](https://x.com/RT_com/status/1917198721973932057) [figures](https://x.com/RT_com/status/1912564244752539838)) received more favorable portrayals. Yves-François Blanchet was largely absent from coverage.
Notably, shortly after assuming the role of prime minister, Carney was targeted in a pre-campaign attempt to discredit his leadership. Russian state media outlet Sputnik published an [article](https://sputnikglobe.com/20250310/top-5-mishaps-of-canadas-new-pm-mark-carney-1121629083.html) falsely accusing him of lacking political and financial competence, engaging in offshore activities, and expressing uncritical support for both climate policies and the Ukrainian "Nazi" regime. RT shared a [photo](https://x.com/RT_com/status/1898938033895793058) depicting Carney alongside Ghislaine Maxwell, a known associate of Jeffrey Epstein. While these images align with a wave of widely circulated AI-generated deepfakes, such as fabricated scenes of Carney with Epstein in a pool or with Maxwell on a beach, the photos posted by RT appear to be authentic, albeit taken over a decade ago. Although the original source of the images is unclear, RT's early publication of them (prior to the official launch of the election campaign) may have contributed to fueling the subsequent spread of deepfakes. Notably, the post attracted significantly more engagement (although still modest) than many of RT's other posts about Carney or the election.
During the election period, RT International published multiple articles and posts portraying Poilievre as "anti-woke," (see this [post](https://x.com/RT_com/status/1904727317378981992) and this [one](https://x.com/RT_com/status/1917198721973932057), for example) echoing his own statements pledging to end what he calls the Liberal Party's "woke agenda" and policies outlined in the Conservative Party's Policy Declaration. This framing aligns with Russian President Vladimir Putin's positioning as an "anti-woke" figure, exemplified by initiatives such as the 2024 "Shared Values Visa," which promotes traditionalist ideals and has been dubbed an "anti-woke visa." RT's amplification of Poilievre's stance could be interpreted as an attempt to appeal to Canada's Russian diaspora and sway their vote in his favour.
Unable to use Meta platforms due to their designation as "extremist" under Russian law, RT also [cited](https://t.me/rt_russian/183032) difficulties in reaching Russian-speaking audiences in the West through Telegram, owing to the platform's regulatory compliance. As a result, RT primarily resorted to posting on X. Yandex/Dzen, a major Russian-language search engine, featured a Canada-specific section for Canadian IPs, but the election-related content there was minimal and general.
Coverage from Russian sources mostly aligned with narratives produced domestically in Canada, including issues such as U.S. [tariffs](https://sputnikglobe.com/20250402/sweeping--relentless-unpacking-trumps-tariffs-1121732482.html) and [annexation threats](https://sputnikglobe.com/20241203/why-isnt-canada-the-51st-state-or-is-it-1121082447.html), [offshore](https://sputnikglobe.com/20250310/top-5-mishaps-of-canadas-new-pm-mark-carney-1121629083.html) finance, the conflict in [Gaza,](https://x.com/RT_com/status/1911350617806692746) and election [debates](https://x.com/RT_com/status/1913027593055727890). Notably, Carney's alleged ties to China were omitted. Instead, more attention was given to Carney's [stance](https://x.com/RT_com/status/1907886082446016521) on the 2022 Freedom Convoy and the alleged [freezing of People's Party members' bank accounts](https://x.com/RT_com/status/1912055128137634198) by the Liberal government. Additionally, [building](https://www.rt.com/russia/613601-zakharova-canada-troops-ukraine/) on narratives portraying Canada as a historical safe haven for Nazis and linking this to its current support for Ukrainian war refugees, Russian state media amplified its criticism of Canada's involvement in Ukraine. Notably, an [X post](https://x.com/RT_com/status/1916926266923716940) published on election day portrayed Liberal-Conservative support for Ukraine as "The Illusion of Choice," a clear effort to meddle in Canadians' voting decisions by influencing or undermining their perception of meaningful electoral options. Additional information about RT's coverage of the election is included in [Section 4.2.4](#page-49-2).
Among Russian-speaking Canadians online, the dominant themes included U.S. influence (tariffs, 51st state), immigration, globalism/woke ideology, Gaza, and elitism. The community also picked up on local concerns such as election fraud claims, the "pen and pencil" voting debate, strategic voting, and Alberta separatism. Post-election commentary was rife with skepticism, with many users expressing disbelief at Liberal victories despite perceived Conservative support ("my family and neighbors all voted Conservative") — paralleling narratives in Ukrainian-speaking communities, where posts also debated strategic voting for Conservative or Liberal.
**Screenshots 3. RT X post published on election day that portrays Liberal-Conservative support for Ukraine as "The Illusion of Choice."**
# **3.2. Misinformation**
Public, media, and policy concern over mis- and disinformation has intensified in recent years, with Commissioner Hogue identifying disinformation as "the single biggest risk to our democracy" in the Foreign Interference Commission's [final](https://foreigninterferencecommission.ca/reports/final-report) [report](https://foreigninterferencecommission.ca/reports/final-report).8 Misinformation is believed to have [circulated widely](https://osf.io/preprints/osf/ubfmx_v1) during the 2021 Canadian election, which occurred amid the COVID-19 pandemic and shortly after the 2020 U.S. election, when widespread election fraud claims culminated in the January 6 insurrection of the Capitol. Canadians perceived there to be a large increase in misinformation between 2021 and 2025, however. In 2021, as measured in the post-election survey, 21% of Canadians believed they had seen, read, or heard political misinformation related to the election. That number leapt to 34% in 2025. This personal experience with misinformation contributed to a large number of Canadians being concerned about its influence on the election. Overall, majorities of Canadians reported being at least moderately concerned, though the intensity varied by party affiliation. Liberals consistently expressed the highest levels of concern, though their worry declined from 87% before the campaign to 76% after the election. Conservatives, by contrast, remained steady at around 70% throughout the period. Among Canadians overall, concern eased slightly, from 78% in the pre-campaign period to 74% afterward. Taken together, these trends suggest that while misinformation was broadly recognized as a problem, the sense of urgency surrounding its disruptive potential diminished after the election, particularly among Liberal supporters.
Throughout the election period, we noted various misinformation claims, such as allegations of compromised foreign connections to political candidates and misrepresentation of policy proposals (see [Section 4](#page-37-1)). Despite several minor and moderate incidents, we assess that misinformation had limited impact and influence during the election.
# **3.3. Bot and inauthentic activity**
Throughout the election period, we monitored for bot and inauthentic activity across social media platforms. We note that, due to serious data limitations imposed by social media platforms on the research and investigation communities, our collective ability to identify bots and inauthentic activity at scale is limited. The [Financial Times](https://www.ft.com/content/743c1dec-0c5c-4c39-a940-830ce156c235), [DFRLabs](https://dfrlab.org/2025/04/25/bot-like-activity-targets-canadian-political-parties-and-their-leaders-ahead-of-election/) and [Reset-](https://thelogic.co/news/canada-election-x-misinformation/)[Tech](https://thelogic.co/news/canada-election-x-misinformation/) all published reports about inauthentic activity online, particularly on X. Beyond what they found, we judge there to be considerably more invisible activity than what we describe here and in [Section 4](#page-37-1). We assessed several potential incidents related to bots during the election and judged the Carney-Epstein association on X to be the most credible. We reported on this incident during the election, but have conducted a more thorough investigation in [Section 4.2.9.](#page-56-1)
# **3.4. Generative artificial intelligence**
Manipulated images, deepfakes, fake news articles, and AI avatars were prevalent and concerning throughout the campaign. [Questions rapidly arose about the potential use of AI](https://theconversation.com/ai-is-making-elections-weird-lessons-from-a-simulated-war-game-exercise-253433) [by political parties](https://theconversation.com/ai-is-making-elections-weird-lessons-from-a-simulated-war-game-exercise-253433) to make their campaign messages more persuasive, sparking debates over whether Canadians are able to reliably detect AI involvement. As shown in Figure 18, Canadians expressed strong concern about AI-generated content misleading the public during elections. Before the election, 84% reported being moderately, very or extremely concerned, with the percentage marginally decreasing to 81% after the election. This pattern and level of concern was fairly consistent across partisan lines. In contrast to other threats, misinformation and foreign influence, Canadians were slightly more concerned about misleading AI content than misinformation in general (79% in the pre-election and 74% in the post-election).
Some of the more disruptive uses of generative AI, detailed in [Section 4](#page-37-1), included fabricated images depicting Carney with Epstein [\(see Section 4.2.9](#page-56-1)), and deepfakes of party leaders circulating in fraudulent social media ads and fake news stories designed to scam Canadians. Beyond these examples, generic deepfakes of politicians (see [Screenshots 3](#page-35-1)) were primarily intended for humour or political commentary, rather than to deceive viewers about their authenticity.
**Figure 18. Percentage of Canadians (by non-partisan and partisan groupings) concerned about the infuence of misleading or depective information during the 2025 Canadian election period (1 month before to 2 weeks after).**
Generative AI is increasingly being used to create entire websites spreading false information. A striking example from the campaign featured AI-generated articles making false allegations about party leaders' financial assets (e.g., falsely claiming that Poilievre has \$25M in assets), which triggered outrage on social media and forced fact-checkers to publish a [correction](https://ici.radio-canada.ca/rci/en/news/2152395/no-pierre-poilievres-net-worth-is-not-25m-despite-what-dubious-ai-generated-articles-say). This use of generative AI to flood the internet with false content is an important source of concern, especially as [AI fuels AI](https://ici.radio-canada.ca/nouvelle/2151457/fortune-investissements-ia-carney-poilievre-desinformation) and
these false statements continued to be propagated by Chatbots even after a correction had been issued.
Overall, while most of the content that circulated during the election was still created by humans, doctored images of party leaders and fake news updates have the potential to mislead segments of the public and to lead to an increasingly polluted information environment.
**Screenshots 4. Examples of AI-generated images that circulated on social media.**
Prime Minister Mark Carney and his wife Diana Fox Carney at Liberal Party headquarters in Ottawa on April 29, 2025.
# **4. Information incidents**
Throughout the course of the election, we detected 35 potential, minor and moderate information incidents (considered as threats) within our information ecosystem. Our detection and analysis revealed a threat landscape where prominent political figures and their respective political parties were the focus of information incidents and the frequent targets of various forms of misleading content and disinformation campaigns. The thematic nature of these threats (see [Table 1](#page-37-2)) revolved around the pervasive character of dis/mis/malinformation, the increasing sophistication of AI-generated or manipulated content, and concerns about foreign influence. Information incidents tend to be aimed at affecting political and electoral processes and manipulating public perception. The severity of these incidents (Level 0: potential incident to Level 3: major incident) escalates with the perceived spread and potential of these themes, moving from potential foreign connections and domestic narratives at Level 0 to direct threats and serious election integrity concerns at Level 3. The framework for incident classification, escalation, and response can be found in the methodology.
The extent the public were aware of these incidents during the electoral period varied. In our post-election survey, we measured Canadians' exposure to events or claims associated with a selected subset of minor (Level 1) and moderate (Level 2) incidents. As shown in [Figure 19,](#page-38-1) the most widely recognized threat was the claim that Canada should become the 51st U.S. state, with four out of five Canadians reported hearing about it. Awareness of other incidents we tracked through the election (evaluated as 'claims') ranged from 23% to 31% of Canadians. Overall, the fact that nearly a third of Canadians reported exposure to claims like urging people to bring their own pen when voting to prevent ballot tampering or images linking Carney to Epstein highlights the relatively widespread reach of misinformation and its coverage during the election.
**Table 1. Information incident themes.**
| Threat Types | Platforms | Countries |
|-------------------------------|-----------|-----------|
| Astroturfing | X | China |
| Bots | Facebook | India |
| Disinformation | TikTok | USA |
| Domestic manipulation | YouTube | Russia |
| Election fraud | Instagram | Ukraine |
| Foreign influence | Bluesky | Vietnam |
| Foreign interference | WeChat | |
| Generative AI/deep fakes | RedNote | |
| Inauthentic platform activity | | |
| Misinformation | | |
The remainder of this section describes all moderate (Level 2), minor (Level 1) and potential information incidents (Level 0) encountered during the election period. Descriptions of moderate and minor incidents include: *Themes* (thematic categorization of the threat type(s), platforms and/or countries involved in the incident outlined in Table 1).
- E *Severity* (justification of why an incident is considered a potential-to-major event).
- E *Motive* and *tactic* (the intended outcome and strategy used).
- E *Origin* (if applicable, where it came from).
- E *Scope and reach* (the extent the incident spread).
- E *Impact* (perceived consequences).
- E *Response* (if applicable, how platforms or other entities responsible responded).
Note, moderate incidents are explained in detail using this structure. Minor incidents embed these elements in the narrative.
**Figure 19. Canadians awareness of a selection of prominent information incidents (claims) during the election.**
Given the volume of potential incidents, our description is mainly summative: We describe recurrent themes and provide examples of threats we encountered during the election. In specific instances we provide additional details of our investigation. [Section 4.1.2](#page-42-1) for example, provides our detailed analysis and findings on the election fraud moderate incident. We also include two spotlight sections within our discussion on minor incidents. These are detailed descriptions of our research and analysis on incidents that received significant media attention (Accusations of Chinese interference on WeChat, [Section 4.2.6,](#page-51-2) and Accusations of Carney-Epstein and bot activity, [Section 4.2.9](#page-56-1)).
# **4.1. Moderate incidents (level 2)**
Two moderate incidents were identified over the course of the election period. As depicted in [Table 2,](#page-38-2) one incident ([docu](https://www.cdmrn.ca/ai-generated-fake-news)[mented extensively during the campaign](https://www.cdmrn.ca/ai-generated-fake-news)) pertains to an organized campaign creating and pushing manipulated and fake ads and content regarding Carney. A second incident relates to persistent and diverse allegations of election fraud which gained traction across the information ecosystem. Both incidents demonstrate a potential to amplify widespread public distrust over the political system and electoral process.
**Table 2. Moderate (Level 2) incidents investigated during the election.**
| Subsection | Incident name | Themes | Description |
|------------|------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------|
| 4.1.1 | Fraudulent political
news content linked to
cryptocurrency scams | Generative AI/deep fakes, dis
information, foreign influence,
foreign interference, Facebook,
U.S., Vietnam, Russia, Ukraine | Sponsored political content with foreign origins,
designed to appear as legitimate Canadian news
sources, and often tied to fake investment schemes |
| 4.1.2 | Surge in claims
undermining election
integrity | Misinformation, disinformation,
domestic manipulation | Widespread dissemination of various election fraud
claims across social media, driven primarily by
right-leaning influencers |
#### **Screenshots 5 (set 1):**
**Examples of ads that circulated on Meta platforms that used AIgenerated websites and images to promote cryptocurrency scams.**
#### **4.1.1. Fraudulent political news content linked to cryptocurrency scams**
**Themes:** Generative AI/deep fakes, disinformation, foreign influence, foreign interference, Facebook, U.S., Vietnam, Russia, Ukraine
The election period was marked by a proliferation of sponsored content masquerading as legitimate news sources (CBC/ Radio-Canada, CTV, Toronto Star, etc.) and generally linked to fake investment platforms. While we observed these ads across social media platforms, Facebook users seemed to be the dominant target.
**Severity:** We classified these fraudulent ads as Moderate (Level 2) incident given the scale (hundreds of ads with tens of thousands of views), sophistication (e.g., use of deepfakes), and the politically divisive nature of the content (frequently invoking combative narratives between parties and politicians),which posed risks to public perceptions of political leaders and confidence in the electoral process.
**Motive:** The main objective was to defraud Canadians.
**Tactic:** The prevailing storyline suggested that a new investment platform had been kept secret by one of the party leaders and
was now revealed to the public. The ads and linked articles often featured deepfakes of Carney promoting the platform or AI-generated images of the party leaders injured or arrested for revealing their secret investment formula. The themes of the ad generally followed the Canadian news cycle, discussing Trump's tariffs at the beginning of the campaign, the leaders' debate in mid-April, and the election results towards the end of the campaign. Some of the content continued to be based on the news cycle after the election, covering King Charles's throne speech, among other things. The fake articles sometimes included plagiarized paragraphs, images, and captions from actual articles published by legitimate media sources. The content was highly sensational and generally supportive of the Liberals. Headlines included language such as "This wasn't just a debate, it was a reckoning. Carney's takedown of Poilievre has the nation buzzing" or presented Carney as "Canada's next PM," prompting many to comment that the election had not yet occurred.
**Origin***:* Identifying the actors behind these fake ads and articles is difficult, given the lack of consistent verification or disclosure of advertiser locations on Meta and the scammers' use of domain registration services intended to hide the identity of the true owners of the fake article websites. In the last week of the
**Screenshots 5 (set 2): Examples of ads that circulated on Meta platforms that used AIgenerated websites and images to promote cryptocurrency scams.**
campaign, we gathered information about the Facebook pages that had not been taken down (platform response described below), showing that a plurality of the pages were created during the election period and that none of the pages appeared to be located in Canada. Pages were primarily located in the U.S., Vietnam, and Ukraine, with [some](https://thelogic.co/news/facebook-canadian-election-scam-political-advertisments/) [evidence](https://www.cbc.ca/news/canada/cbc-ads-fake-investment-scheme-1.7439812) pointing to Russian involvement. In addition to newly created pages, more prominent pages, such as those of former U.S. State Representative Renitta Shannon and Spanish vinyl sticker company La Tienda de las Pegatinas, were also hacked and used to promote these scams in Canada. The bulk creation of pages using similar visuals and publishing the exact same ads, as well as the recurring use of the same pages to spread similar scams across multiple countries, further highlights the international nature of this scheme and is indicative of a coordinated strategy. Our investigation of the links included in the ads reveals that domains served different content depending on whether the website visitor's internet address was Canadian, with only Canadian visitors being shown a fake article promoting the fraudulent investment platform. The majority of the IP addresses originated from California.
**Scope and reach**: This widespread network of coordinated pages and websites engaged a large Canadian audience during the election and received extensive media coverage nationally [e.g., [1](https://thelogic.co/news/facebook-canadian-election-scam-political-advertisments/), [2](https://www.ledevoir.com/politique/canada/872032/reseaux-sociaux-sont-inondes-contenu-election-cree-ia), [3](https://www.journaldemontreal.com/2025/04/15/publicites-frauduleuses-sur-facebook-meta-autorise-des-fausses-nouvelles-sur-les-elections-canadiennes), [4](https://www.cbc.ca/news/canada/more-fake-cbc-ads-investigation-1.7494923), [5](https://thelogic.co/news/facebook-deepfakes-mark-carney-canada-election/)] and internationally [e.g., [1](https://www.nytimes.com/2025/04/21/technology/canada-election-facebook-instagram-meta.html?smid=nytcore-ios-share&referringSource=articleShare), [2](https://www.theguardian.com/world/2025/apr/18/canada-fake-political-content-social-media)]. Based on our team's
platform monitoring, tips from the public (using our tipline), and keyword searches on social media platforms, we identified 100 Facebook pages promoting these ads during the election and more than a hundred additional pages in the month after the election. However, the pages we investigated are not exhaustive of the content circulating and are [likely to only represent the](https://www.reset.tech/resources/dormant-danger-research-report/) [tip of the iceberg](https://www.reset.tech/resources/dormant-danger-research-report/). Most of these ads did not appear in Meta's Ad Library and, when they did, were not self-declared as political, making it difficult to precisely estimate how much was spent and how many Canadians were exposed. However, our analysis of the ads that were labelled as political in the Ad Library suggests that tens of thousands of dollars or more were spent on these ads during the campaign. In our survey, 24% of Canadians indicated that they "encountered any social media post or webpage that falsely presented itself as a legitimate news source (e.g., CBC, Toronto Sun, La Presse) by imitating its name, logo, or design" over the past month, with slightly more than half (52%) indicating that they did not encounter such content and 1 in 4 (24%) being unsure. The ad we observed with the largest reach generated more than 6,000 interactions (reactions, comments, reshares), which, based on [studies with privileged](https://doi.org/10.1126/science.abp9364) [access to Facebook data](https://doi.org/10.1126/science.abp9364), suggests that it was viewed by more than 100,000 Canadians. Similar ads also circulated on X and YouTube, although we encountered them far less frequently and received significantly fewer crowdsourced submissions via the tipline compared to Facebook.
**Screenshots 6: Disclosure of AI-generated content in ads on Meta, from [Meta.](https://www.facebook.com/government-nonprofits/blog/political-ads-ai-disclosure-policy)**
**Impact:** A majority of Canadians appeared to be able to recognize the content as fake, limiting disruptive effects on the election. Indeed, 58% of survey respondents who believed they were exposed indicated that they immediately recognized the content as fake, while 42% only recognized it later. It is important to note that people cannot report exposure if they did not recognize it. However, the 24% who are unsure whether they were exposed likely include individuals who saw the content without realizing it was fake. Additionally, our analysis of 770 comments posted on the fake ads shows that about three-quarters of them flagged the content as fake. The remaining comments were generally critical of Carney or the CBC. Given Canadians' relatively high ability to identify the content as inauthentic and the fact that [election ads generally have minimal effects on vote choice,](https://journals.sagepub.com/doi/full/10.1177/20531680221076901) the ads are unlikely to have had any significant impact on the election results. However, they might have shaped broader attitudes, including perceptions of information manipulation and trust in the election. For example, in a video that gained significant traction, a prominent influencer claimed the ads contravened election regulations and represented a form of foreign interference with the potential to affect the election's outcome. This type of fraudulent content also [poses a risk for](https://journals.sagepub.com/doi/10.1177/2056305120903408) [public trust in news media brands](https://journals.sagepub.com/doi/10.1177/2056305120903408) and can increase cynicism and disengagement from the broader information environment.
**Response:** While Meta took down many of the ads and pages, their response was largely reactive and intermittent. In particular, the use of automated tools to identify political ads or AI-generated deepfakes of political leaders appeared limited or even absent during the election. Most fraudulent ads featuring political leaders were not labeled as political, removals appeared ad hoc, and deepfakes from removed ads were often reposted several times and not immediately taken down. More than a third of the pages we identified had not been removed by the end of the campaign, and removal of specific ads and pages did not prevent new pages from being created and reposting the exact same content the next day. Fraudulent ads remain as prevalent on the platform as they were during the election. In order to limit the disruptive impacts of AI, Meta introduced [new](https://www.facebook.com/government-nonprofits/blog/political-ads-ai-disclosure-policy) [measures](https://www.facebook.com/government-nonprofits/blog/political-ads-ai-disclosure-policy) in January 2025 requiring advertisers to self-disclose their use of digitally created or altered media when buying social issue, electoral, or political ads. Meta then adds a 'Digitally created' label next to the payer's name in the ad disclaimer (see Screenshots 5 ([1 of 2\)](#page-39-2) and ([2 of 2](#page-40-0))).
Fraudulent ads including deepfakes of political leaders generally did not include such labels related to AI use. This type of self-declaratory approach is likely to be particularly ineffective for harmful content like financial fraud, where advertisers deliberately seek to convince the audience of the authenticity
of the content and may already anticipate that their ads will be removed for violating platform policies. Meta is currently facing a [class action lawsuit](https://ici.radio-canada.ca/nouvelle/2169582/fraude-crypto-action-collective-facebook-meta) regarding fraudulent ads on the platform. Additional information on this incident can be found in the [two](https://www.cdmrn.ca/ai-generated-fake-news) [incident response reports](https://www.cdmrn.ca/ai-generated-fake-news) published during the election period.
#### **4.1.2. Surge in claims undermining election integrity**
**Themes:** Domestic manipulation, misinformation, disinformation
As described briefly in [Section 3.2,](#page-35-2) an increasing volume and severity of concerns over the integrity of the election grew as election day came closer. Questions around the security of the ballot (Elections Canada providing pencils and not pens, insinuating that a voter's ballot could easily be changed) and Liberal-partisan influence over Elections Canada grew during advanced polling dates. Types of election fraud varied widely with 7 distinct claims emerging: Voter ID insecurity, ballot manipulation, bring your own pen, hashtag "too big to rig", fake polls, unfairness of the long ballot, and the Terrebonne recount.
**Severity:** This was a moderate incident given the significant scale (at least 500,000 online engagement), broad offline reach (28% of survey respondents reporting exposure), and potential effect on trust in Canada's electoral institutions.
**Motive:** This was largely a domestic conversation to cast doubt on the integrity of the election, a trend consistent with similar claims emerging during other democratic elections in Canada (e.g., [Elections BC](https://meo.ca/work/british-columbia-election-information-ecosystem-project)) and around the world. Casting doubt on election integrity can be a political tool but may also reflect genuine concern about the integrity of the vote.
**Tactic:** Trending topics and engagement were mostly driven by influencers and partisan supporters who amplified unverified anecdotes and suspicions around the election process.
**Origin**: There was no single source of election fraud claims. This incident was comprised of numerous claims that called the integrity of the election into doubt at different times throughout the election period and threatened to (unjustifiably) reduce trust in the election process. Some of these claims were from individuals who reported on their personal anecdotes of what they perceived to be election-compromised processes, which were then picked up and amplified by influencers (e.g., ballot manipulation and voter identity claims). Others originated from the influencers themselves (e.g., hashtags like "too big to rig"). Influencers may also have benefited from the amplification of this content which gained increasing interest over time.
**Scope and reach***:* We found that one in four Canadians were exposed to claims of election fraud during the campaign. We
identified 1,533 posts related to fraud claims from our core dataset of influential Canadian voices, which garnered over half a million likes ( approximately 523,000). This is an underestimate of the scale of the issue since our dataset only includes Canadian influential accounts that meet a follower threshold (at least 10K for X and TikTok, or at least 5K for Instagram, Youtube, Telegram, and Bluesky) and the search was limited to keyword matches (explained further in the Methodology section). This incident had a wide scope and reach.
**Impact:** The prevalence of election fraud claims coincides with a small drop in trust in Elections Canada among Conservative supporters (54% to 61%) from before to after the election. More broadly, the rise of election fraud claims coincides with declining trust in election management bodies in many [democracies](https://www.reuters.com/world/global-index-free-fair-elections-suffers-biggest-decline-record-2023-democracy-2024-09-17/?utm_source=chatgpt.com) [around the world](https://www.reuters.com/world/global-index-free-fair-elections-suffers-biggest-decline-record-2023-democracy-2024-09-17/?utm_source=chatgpt.com).
**Response:** Elections Canada made several posts and articles online to combat major claims of election fraud, such as explaining [why pencils are provided](https://x.com/ElectionsCan_E/status/1920525200329707529) in polling stations in response to the "bring your own pen" narrative. This demonstrates the severity of claims to disrupt electoral processes as well as responsive institutional engagement to address electoral security. However, these responses often received much less visibility than the original claims of fraud. In the case of poll workers being accused of taking unsealed ballots home, some of the [most liked](https://x.com/Inevitablewest/status/1916898833180545316) [posts](https://x.com/Inevitablewest/status/1916898833180545316) received over 2.4M views compared to Election Canada's [response post](https://x.com/ElectionsCan_E/status/1916853602720563702) receiving only around 170,000 likes.
Given that this incident emerged with prominence late in the campaign, we did not publish full findings in regular incident updates. The analysis below explains and concludes our detailed investigation of this incident.
# **SPOTLIGHT**
We classify election fraud as a case of *misinformation* rather than *disinformation* as everyday individuals may share false or manipulated claims of election fraud due to genuine concern over the security of the election process and importance of election security. However, we do not dismiss that the creation and amplification of these claims may also be disingenuous and intentional in nature, by actors who may benefit from polarized and engagement attention online or greater distrust in the electoral system.
Various claims of election fraud spread to a substantive degree during and after the election period, with 28% of Canadians reporting having heard of the claim that "the federal election was rigged or fraudulent." Of the various claims of fraud, 29%
**Screenshots 7: Examples of influential voices sharing election fraud claims.**
of Canadians were aware of the claim urging voters to "bring your own pen to vote so your ballot won't be altered," for example. We used social media data analysis, surveys, and avatar based monitoring methods to understand the dissemination and impact of election fraud claims made online.
The main election fraud claims identified through avatar-based monitoring and the tipline were:
- E **Voter Identity:** Voter ID insecurity leading to fraudulent votes claims that non-electors were able to vote, that individuals went to vote but found their name had already been crossed out or that they were not registered, and reports of people voting at multiple polling stations.
- E **Ballot Manipulation:** This focused on claims of Elections Canada failures to maintain a secure chain of custody of the ballots. These included claims of Elections Canada throwing away ballots, election workers taking unsealed ballot boxes home, insecure handling of ballots at polling stations, discoveries of ballots in dumpsters, and claims that elections workers were attempting to persuade voters at the polling station
- E **Pens not Pencils:** This bring your own pen was a commonly repeated call for electors to bring their own pens instead of using the provided pencil so their vote can't be erased or changed by Elections Canada workers.
- E**"Too Big to Rig":** This focused on the claim that the election was likely to be rigged in favor of one political party (commonly the Liberals) if another party (typically the Conservatives) do not have a large enough margin. The claim here is that election fraud will occur but enough support will en-
sure the fraud is unsuccessful. This is a powerful politically mobilizing call but also necessarily focuses on fraud.
- E **Fake Polls**: An increasingly common phenomenon in Canada and question the validity of polls and claim that pollsters are trying to manipulate public opinion in favour of one party or another. We observed the use of generative AI content to discredit polling firms as well.
- E **Long Ballot:** This focused on the Carleton riding and the efforts of the Longest Ballot Committee to put 91 candidates on the ballot. Claims focus on Elections Canada allowing or even encouraging this lengthy ballot to help defeat Poilievre. We observed a similar initiative and similar claims for the by-election in Battle River-Crowfoot where over 200 candidates were registered and Elections Canada implemented a **[write-in ballot](https://www.cbc.ca/news/politics/elections-canada-longest-ballot-1.7595763)**.
- E **Terrebonne Recount:** This saw the Liberal Party candidate defeat the incumbent by a single vote. Claims abounded that Elections Canada fraudulently awarded this riding to the Liberals after a recount.
We additionally carried out automated discovery of fraud claims using an LLM to further confirm the list of main claims. This analysis found similar claims listed above, as well as more granular claims such as mail-in ballot controversies, voter suppression claims, claims of transparency issues surrounding Elections Canada, and general claims of rigged ballot counting (e.g., "phantom math" or the artificial manipulation of vote counts or turnout rates). For simplicity, our analysis focuses on the larger main claims listed above.
In our social media dataset, we were able to identify (using
**Figure 20. Daily volume of posts claiming election fraud from prominent Canadian voices online, with both daily post count and number of daily likes shown, for the election period (Feb. 23 and May 28, 2025).**
a combination of a dictionary and LLM approach, see **[Meth](#page-65-1)[odology](#page-65-1)** for more details) 1,533 posts from Canadian news outlets, politicians, and prominent influencers that claimed election fraud. These posts, in total, received around half a million (538,390) likes during the period between Feb. 23 and May 28. The posts were collected from X, Instagram, TikTok, Youtube, Telegram, and Bluesky.9
**[Figure 20](#page-44-0)** shows the daily post volume trend and the daily engagement (measured by likes) volume trend of election fraud claims during the election period. Overall, we see that discussion around election fraud increased steadily since the issuance of the writ, peaked one day after the election, and declined rapidly afterwards. By May 28, engagement around election fraud claims was negligible among influential Canadian entities.
Individual election fraud claims emerged and waned at different times during the election period. Depicted in **[Figure 21](#page-45-0)**, prior to election day, popular claims included raising doubt on national polls (*Fake Polls*), calling on Conservative voters to make the election "too big to rig" (*Too Big to Rig*), and encouraging voters to use a pen instead of a pencil to secure their votes (*Pens Not Pencils*). In contrast, the fraud claim topics discussed immediately after the election included accusations that Poilievre lost the Carleton riding due to Elections Canada unfairly
allowing its long ballot (*Long Ballot*), ballot manipulation, and ineligible persons voting (*Voter Identity*). The last election fraud claim to emerge was one claiming that Terrebonne's recount wrongly awarded the riding to the Liberal party (*Terrebonne Recount*). This emerged around May 11, when Terrebonne's recount results were announced and slowed down around May 20. Despite the number of claims, none of them sustained long term or organized interest in the Canadian media ecosystem.
When we compare election fraud claims between social media platforms, we see a disparity between how posts are distributed and how engagement is distributed. When it comes to post volume, the vast majority (between 79% and 94%) of election fraud claims (labelled as percent of posts by platform per issue) were posted on X (see **[Figure 22](#page-46-0)**). However, engagement with election fraud claims (labelled as percent of likes by platform by issue) was primarily distributed across YouTube (47%), X (45%), and TikTok (6%). While general election fraud content (claiming fraud without specifying an issue) received the most engagement on X, topics of ballot manipulation, fake polls, the long ballot initiative, "too big to rig," and voter identity received the majority of their engagement from YouTube. Meanwhile, the topics of using a pen to vote and Terrebonne recount received the most engagement from TikTok. Therefore, although election fraud claims may have been easiest to post
**Over-time engagement with seven prominent election fraud claims from Feb. 23 and May 28, 2025.**
on a text-based platform like X, video-based platforms like YouTube and TikTok allowed them to gain traction.
Posts making election fraud claims were dominated by influencers in contrast to politicians, news outlets, and civil/government organizations. They had the highest volume of election fraud posts (90% of all fraud posts), and also received the most engagement on that content (85% of all engagement on fraud posts). The top five influencers, all right-wing, in our seedlist received 59% of all engagement received by posts claiming election fraud. Very few news outlets made claims of election fraud; among the few, True North received 88% of all engagement on election fraud claims coming from news outlets. Among politicians, PPC candidates posted 95% of content claiming election fraud and received 98% of the engagement on election fraud claims.10 Politicians belonging to major national parties rarely implied or claimed election fraud (Conservative 4%, Liberal 0.5%, NDP 0.5%). It is important to note that politicians as a whole received very little engagement on election fraud claims they made: at 5,508 likes total, politicians account for 1% of engagement received by fraud claims from the entire seedlist. Regarding foreign sources of election fraud claims, there were
a **[few](https://x.com/Inevitablewest/status/1916898833180545316) [election fraud](https://x.com/DerrickEvans4WV/status/1917578072011743241) [posts](https://x.com/JimFergusonUK/status/1917356399022981278)** from non-Canadians that achieved a high level of engagement. However, non-Canadian entities did not appear to be major or persisting actors in spreading election fraud claims online.
Our avatar-based observations revealed that conservative social media accounts were the most heavily exposed to election fraud content, particularly on X, Instagram, and Facebook. Centrist or apolitical accounts were less subject to fraud claims, while still exposed to the popular posts, especially on Tiktok, Instagram, and X. Left-leaning accounts received much less fraud content, especially on Instagram and Facebook.
Immediately following the election, ecosystem monitoring remained critical in order to capture any lingering or emergent threats related to the election. This period was marked by a surge in misinformation and disinformation, particularly narratives questioning the legitimacy of the electoral process and Elections Canada. These included unsubstantiated claims of voter fraud, gerrymandering, foreign interference, and media bias, with notable incidents such as false accusations of manipulation in Milton East-Halton Hills South, allegations of rigging in Pierre Poilievre's riding, and disputes over results in Terrebonne.
**Figure 22 Comparison of perceived fairness of election administration for 2021 versus 2025 elections by partisan identification.**
**Figure 23. Comparison of perceived fairness of election administration for 2021 versus 2025 elections by partisan identification.**
**Figure 24. Trust in Elections Canada in the 2025 pre-election and post-election periods by partisanship.**
The prevalence and reach of claims of election fraud online could undermine confidence in the impartiality of electoral administration and Canadian democracy more generally. Given the central role of Elections Canada in safeguarding democratic legitimacy, even modest declines in trust may have serious consequences for perceptions of fairness and acceptance of election results. **[Figure 23](#page-46-1)** provides our best evaluation of whether or not the election or these claims impacted population-level attitudes. We find mixed results.
Trust in Elections Canada (see **[figure 24](#page-47-1)**) remains very high overall, with nearly four in five Canadians expressing confidence by election day. However, partisan divides are evident and widening. Among Liberals, trust is nearly universal and increases during the campaign (from 89% to 94%). Among Conservatives, trust starts considerably lower (64%) and declines slightly by the post-election period (61%). These differences align closely with patterns in the online information environment, where right-wing social media users were disproportionately exposed to election fraud narratives. Taken together, the findings suggest that while Canadians as a whole continue to trust electoral institutions, misinformation and claims of election fraud may have selectively eroded confidence within the Conservative base.
# **4.2. Minor incidents (level 1)**
Twelve minor incidents (Level 1) were identified during the election period. Foreign influence narratives and campaigns were particularly evident, making up seven out of the twelve incidents. Chinese and American foreign influence concerns were at the forefront, including allegations of suspicious economic ties between Carney and China, pro-Liberal Party campaigns on WeChat, pro-Conservative Party campaigns on RedNote, and U.S.-Canada advocacy connections for the 51st statehood narrative. AI-generated or manipulated content was also an emerging concern, largely linked to the discrediting of Mark Carney. We noted reports of AI-generated content linking Carney to Epstein and deep-fake images of Carney. AI-generated avatars on X were also of concern.
**Table 3. Minor (Level 1) incidents investigated during the election.**
| Section | Incident name | Themes | Description |
|---------|--------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| 4.2.1 | Suspicious 51st state
Facebook groups | Facebook, foreign influence, do
mestic manipulation, astroturfing,
inauthentic platform activity, U.S. | Astroturfing campaigns on Facebook promoting
Canadian annexation as the 51st U.S. state using
repurposed groups to simulate grassroots support |
| 4.2.2 | Suspect American influence
in campaign advocating for
annexation of Alberta | Foreign influence, domestic ma
nipulation, inauthentic platform
activity, U.S. | Campaign promoting Alberta's annexation as the
51st U.S. state, backed by U.S. media, influencer net
works, and coordinated offline mobilization |
| 4.2.3 | Suspicious U.Sbased
recruitment of Canadians | Foreign influence, U.S. | Anonymous job listing targeting right-leaning
Gen-Z Canadians with vague political tasks, raising
concerns over covert foreign influence |
| 4.2.4 | Suspicious amplification
of anti-liberal narratives
between RT and conservative
outlets | Foreign influence, X, Russia, U.S. | Russian state media amplification of anti-Liberal
and polarizing narratives, possibly in conjunction
with Canadian right-wing outlets or actors support
ive of MAGA-style politics |
| 4.2.5 | Accusations of economic ties
between Mark Carney and
China | Foreign influence, disinformation,
information manipulation, China | Social media campaign portraying Mark Carney as
tied to the Chinese government, using out-of-con
text or edited footage |
| 4.2.6 | Alleged Chinese Pro-Liberal
Influence on WeChat | Foreign influence, WeChat, China | Allegations of Chinese interference focused on We
Chat activity targeting diaspora communities with
content supportive of Carney |
| 4.2.7 | Alleged Chinese pro-Con
servative Influence on Red
Note | Foreign influence, bots, RedNote,
China | Newly created RedNote accounts targeting Chi
nese-speaking Canadians with pro-Conservative
content, showing signs of coordinated and in
authentic engagement |
| 4.2.8 | Surge in deep fakes of polit
icians | Generative AI/deepfakes, disinfor
mation, TikTok | Proliferation of AI-generated political content on
social media, including deepfakes of politicians and
faceless accounts |
| 4.2.9 | High engagement in Mark
Carney-Jeffrey Epstein false
claims | Generative AI/deep fakes, disinfor
mation, bots, X | False narratives linking Carney to Epstein, spread
through AI-generated images and coordinated
posts on social media |
| 4.2.10 | AI generated avatars on X
sharing partisan content | Bots, Disinformation, X | Anonymous X accounts with AI-generated profile
pictures sharing high volumes of anti-Liberal and
pro-Trump content |
| 4.2.11 | Anti-Carney and liberal party
networks on X | Bots, generative AI/deepfakes, mis
information, X | Small networks of inauthentic X accounts promoting
anti-Carney messages |
| 4.2.12 | Anti-conservative button
planting and election integrity
concerns | Domestic manipulation, misinfor
mation | Covertly planted buttons by Liberal staffers at a
Conservative event spark widespread engagement
and backlash |
**Screenshots 8. Membership application form for the repurposed 'Canadians for the 51st State and Elon Musk for Governor' Facebook group.**
### **4.2.1. Suspicious 51st state Facebook groups**
**Themes**: Facebook, foreign influence, domestic manipulation, astroturfing, inauthentic platform activity, U.S.
51st State Facebook Groups involving coordinated [astroturf](https://www.cdmrn.ca/learning-content/astroturfing)[ing campaigns](https://www.cdmrn.ca/learning-content/astroturfing) (multiple dormant Facebook groups and pages promoting statehood in various provinces were reactivated and sustained by inauthentic accounts) were identified during the election to artificially amplify support for Canada's annexation as the 51st U.S. state. This incident was escalated due to these groups having large membership and high prominence in search results when searching for 51st statehood in Canada, and their covert nature. We were unable to identify a single entity behind these groups and it seemed to be a range of Facebook users.
Emerging amid heightened U.S.-Canada tensions over the threat of tariffs and calls to contest Canadian statehood, this incident repurposed largely dormant Facebook groups to amplify narratives pushing for Canada to become the 51st state. For example, groups such as — the "Canadians for the 51st State and Elon Musk for Governor" Facebook group (see [Screenshot](#page-49-3) [8\)](#page-49-3) simulated public engagement and interest in the matter. The reach of this incident remained limited, as Canadian users were quick to identify and publicly call out the inauthentic nature of these groups, leading to widespread skepticism of its grassroots nature. Additionally, mainstream media coverage, including a detailed article by the [Toronto Star](https://www.thestar.com/news/canada/the-story-of-canadians-for-the-51st-state-and-elon-musk-for-governor-how-a/article_7e8526dc-72ae-49b5-81e1-297619275fed.html), played a significant role in informing the public about the repurposing of this page and curbing the campaign's influence. This incident was closed in early April after concluding that the repurposing of Facebook groups appeared to be isolated, with no evidence of coordinated or networked malicious activity or paid promotion through Facebook's advertising systems, and notably, the activity stemmed primarily from within Canada. It is important to note, however, that there may have been additional astroturfing campaigns operating on Facebook during this period that were not identified or confirmed as part of this investigation.
#### **4.2.2. Suspect American influence in campaign advocating for annexation of Alberta as the 51st US state**
**Themes:** Foreign influence, domestic manipulation, astroturfing, inauthentic platform activity, U.S.
Advocacy campaigns promoting Alberta's annexation as the 51st U.S. state arose during the end of the electoral period. While this narrative was first detected in February 2025, it intensified into physical mobilization efforts in May during the post-Federal election period. This was classified as a minor incident as the campaign raised significant concerns due to its transnational ties and polarizing effects, especially in view of a referendum to be held in Alberta in [2026](https://ici.radio-canada.ca/rci/en/news/2163229/alberta-premier-danielle-smith-defends-promise-of-a-potential-referendum-on-separation).11 The movement blended both genuine grassroots activism and interest with astroturfing by organized entities to manufacture and distort widespread public support. Spearheaded by two key entities, the America Fund, and the Alberta Prosperity Project (APP), campaigning both online and offline occurred. These campaigns included influencer partnerships, petition drives, billboard campaigns, and [conversations with officials](https://www.desmog.com/2025/07/22/trump-officials-discussed-500m-alberta-independence-loan-separatist-claims/?utm_source=chatgpt.com) in Washington, D.C. These actions were amplified by right-wing U.S. media, with endorsements
**Screenshots 9. Billboard promoting Alberta separation, sponsored by America Fund.**
**Figure 25. Evaluation of future options for Alberta among Albertans in the pre- and post-election periods.**
from outlets such as Fox News. The movement's momentum was fueled by an online petition, increased media visibility, and events timed with the G7 summit.
While we observed increased discussion around the separation and annexation of Alberta during the post-election period, we did not observe (through our survey data) a significant increase in the percentage of organized mobilization of Albertans supporting independence from Canada. Our survey findings suggest that, when given the choice, most Albertans would prefer a revised federal arrangement in which the province has more autonomy than complete separation from Canada (see [figure 25](#page-50-3)).
#### **4.2.3. Suspicious U.S.-based recruitment of Canadians**
**Themes:** Foreign influence, U.S.
On March 24, an anonymous U.S.-based company posted a high-paying, vague job listing on Upwork (a freelance job platform) seeking to recruit right-wing Gen-Z Canadians for an unidentified job. Follow-up posts by the same company specifically sought "culturally-connected" individuals to explain anti-American sentiments. The listing offered USD 300 for a low-effort task and provided no clear information about the company, the purpose of the job, or how the material would be used. Given concern about the potential for covert foreign influence, vague objectives, and potential to manipulate public discourse, this event was classified as an incident. It was considered minor (Level 1), however, given limited reach. While intent remains unclear, the selective recruitment and follow-up listings suggest a covert effort to gather partisan narratives or
stoke division under the guise of research. The listing quickly drew scrutiny across Facebook, Threads, and other platforms, particularly among politically engaged communities in Canada (See [Screenshots 8\)](#page-49-3). While it did not appear to directly interfere with the electoral process, it contributed to growing unease over foreign influence. Upwork removed the listing by April 5, though without explanation, and no major platform interventions occurred beyond user reporting and grassroots awareness-raising.
### **4.2.4. Suspicious amplification of anti-liberal narratives between RT and right-wing outlets**
**Themes:** Foreign influence, X, Russia, U.S.
Russian state media outlet RT amplified polarizing and anti-Liberal Party narratives during the Canadian election period, often in alignment with Canadian right-wing outlets and MAGAaligned actors. While this activity had the potential to be a case of significant foreign influence, this incident was considered minor given limited volume and impact. Throughout the campaign, RT's motive appeared to be undermining the Liberal Party and advancing skepticism toward Canadian democratic legitimacy, using tactics such as highlighting Poilievre's popularity, mocking Carney's debate performance, spreading allegations of elitism and ties to Epstein, and surfacing divisive themes including globalization, gender inclusivity, immigration, Israel, and the Freedom Convoy.
RT's coverage was echoed by Canadian MAGA supporters with large social media followings and content was often shared live
**Screenshots 10. Facebook and X posts raising concerns of foreign interference over the Upwork Job posting.**
or in close succession across multiple platforms, amplifying its visibility and resonance. The result was the spread of divisive political discourse, contributing to skepticism about democratic legitimacy and fostering polarization on key issues; even absent confirmed coordination, the efforts of Russian state media can be seen as an attempt to exert foreign influence over Canadian electoral politics. The amplification pattern suggests deliberate monitoring and attempts at agenda-setting by RT.
#### **4.2.5.Accusations of economic ties between Mark Carney and China**
**Themes:** Foreign influence, disinformation, information manipulation, China
Since late March 2025, social media posts, headlines, and video clips circulated widely accusing Carney of ties to the Chinese government. Importantly, most posts relied exclusively on dated and out-of-context footage. This incident was considered Level 1, as the claims lacked verifiable evidence but posed reputational risks to Carney during the federal election by reinforcing foreign interference fears. The likely motive was to erode public trust in Carney and the Liberal Party by portraying him as compromised or aligned with the Chinese Communist Party. Key tactics included resurfacing old speeches without context, quoting edited video clips,employing emotionally charged hashtags like #CantoneseCarney, and spreading AI-generated memes (e.g., Screenshots 10). The narrative began gaining traction around
March 25 after a [National Post article](https://nationalpost.com/news/canada/pierre-poilievre-mark-carney-china-ties) which reported the accusation from the Conservative leader that Carney's previous roles at Brookfield Asset Management may have exposed him to ties with China. Since then, the accusation spread across X, Reddit, Facebook, and Instagram, fueled by right-leaning influencer accounts and politicians. While engagement was concentrated in partisan spaces, the campaign persisted across platforms.
The impact was a heightened level of public suspicion about Chinese influence and interference and distraction from evidence-based foreign interference discussions (ie, those grounded in verified intelligence findings or flagged by government agencies like the SITE Task Force). No major platform intervention occurred, though Carney publicly denied the allegations and several mainstream outlets issued fact-checks and clarifications.
#### **4.2.6. Alleged Chinese pro-Liberal influence on WeChat**
**Themes:** Foreign influence, WeChat, China
Allegations of pro-Liberal Chinese foreign interference circulated widely during the election. Concerns were raised following an [investigation](https://www.canada.ca/en/privy-council/news/2025/04/information-operation-on-wechat-targeting-the-45th-general-election.html) by the SITE Task Force into suspected foreign interference involving a China-based WeChat channel's coverage of the Canadian federal election, see Screenshots 11. This incident was classified as minor (Level 1) given the significant risk to the integrity of the election of Chinese interference. The
**Screenshots 11. Repost of a video by RT using similar text, shared shortly after the original Post Millennial upload.**
**Screenshots 12. Posts accusing Carney of holding ties with Beijing.**
Chinese government's likely motive was to foster a favorable view of Liberal leadership, particularly Carney, within Chinese-Canadian communities to support stable bilateral relations. The tactic, as flagged by the SITE report, appeared to be a perceived editorial bias in support of Carney, citing laudatory language in reference to him on WeChat. The SITE Task Force concluded that the channel's information operation aimed to influence Chinese communities in Canada in the context of the 45th General Election. Further monitoring and analysis of the WeChat channel *Youlieryoumian* (flagged by SITE) yielded less concerning results. We independently assessed this channel's activity (see Spotlight) and found no cause for alarm or sign
that China has materially interfered in the Canadian election using this channel. Notably, we found that Canada-related content was very limited compared to a dominant focus on the U.S.-related themes such as Trump's tariffs and foreign policy. Only a few posts addressed Canadian politics, see [Screenshots](#page-52-0) [11\)](#page-52-0), ["The U.S. Has Met Canada's Tough Prime Minister](https://freewechat.com/a/Mzg3MjEyMTYyNg==/2247656158/1)" (dated March 10, 2025) and "[Canada's Road to a 'Survival Election](https://freewechat.com/a/Mzg3MjEyMTYyNg==/2247656809/1.)'" (dated March 25, 2025). One of their most recent posts on May 7, 2025 reported on Carney's first visit to the White House as prime minister. No apparent misleading narratives or unusual activity aimed at influencing Canadian public opinion were observed on this channel.
#### **Screenshots 13:**
**Two articles on Canadian politics posted by the WeChat channel Youlieryoumian, flagged by SITE. The article on the left is titled "The US has met Canada's tough Prime Minister," dated March 10, 2025. The original article is available [here](https://freewechat.com/a/Mzg3MjEyMTYyNg==/2247656158/1). The article on the right is titled "Canada's Road to a 'Survival Election'," dated March 25, 2025. The original article is available [here](https://freewechat.com/a/Mzg3MjEyMTYyNg==/2247656809/1.).**
# **SPOTLIGHT**
There has been notable concern regarding the activity of Youli-Youmian during the election, including from many Members of Parliament and candidates. Conservative candidate Michael Chong, **[previously targeted by Chinese foreign interference](https://www.cbc.ca/news/politics/wechat-disinformation-operation-chong-1.6931377), [responded with a statement](https://x.com/MichaelChongMP/status/1909326721087664402/photo/1)** alleging China is spreading information to support the Liberal Party. Other responses claim Carney is financially beholden and compromised to the government of China. These claims received widespread attention in both traditional and social media, with continued attention on foreign interference in the campaign. First, we note that the intelligence flagged by SITE that links the ownership and editorial focus of the channel to the People's Republic of China Chinese Communist Party's Central Political and Legal Affairs Commission should be integrated into international lists of state-linked entities (e.g., EEAS SG.Strat).
For our investigation we reviewed all content posted by the channel in the months of March and April. We extracted their engagement metrics and country of focus as well as evaluated their contents and editorial position. We asked: 1) Did the content of the channel disproportionately focus on Canada?, 2) Was the content inconsistent with the existing editorial approach of the channel?, 3) Did the content focused on Canada receive disproportionate engagement?, and 4) Is there any evidence of impact?
An analysis of the Youli-Youmian channel's content shows that Canada-related coverage was minimal: Of 93 posts on the channel in March 2025 only 4 posts (4.3%) of in March 2025 focused on Canada. This is compared to 58.1% that focused on the United States, along with substantial coverage of Ukraine (15.1%), Korea (7.5%), Taiwan (6.5%), and the EU (5.4%). Canada was not a central focus for the channel or its audience during March 2025. The Canadian election was likely covered due to its significance as a major international event, rather than the result of a targeted influence campaign.
The channel focuses on providing information on international affairs with a strong and critical focus on the United States. In the articles focused on Canada, there is no clear evidence of disinformation, misleading narratives, or manipulative framing. The language used appears factual and does not clearly favor any particular party or candidate. The comments allowed to be posted under the articles do not favour one party or another. Based solely on the textual content, we find little evidence to suggest an intent to influence political outcomes in Canada.
**Figure 26. Normal pattern of engagement with Canadian-focus stories on Youli-Youmian WeChat channel.**
Of particular concern was the volume of engagement on these posts, reported as higher engagement than that received by posts of major state media outlets such as the People's Daily. **[Figure 26](#page-54-2)** shows the four Canadian stories along the distribution of engagement for all March posts on the channel. When compared with other posts published during the same period, the Canadian content did not attract unusually high engagement, and overall level of engagement was consistent with the channel's popularity. The post with the most engagement (March 10) was in the 77th percentile for engagement: it received more engagement than 71 other posts but less than 21 other posts. Engagement with the Canadian stories was not, on average, higher than that of these other countries and was significantly lower than Korea, Ukraine, and the United States. There is no indication that these posts stood out in ways that would suggest targeted amplification or exceptional audience interest.
#### **4.2.7. Alleged Chinese pro-Conservative influence on RedNote**
**Themes:** Foreign influence, RedNote, China
At the beginning of the election period, several newly created accounts emerged posting exclusively pro-Conservative Party content targeting Chinese-speaking communities on RedNote12, a popular Chinese lifestyle social media platform. The incident
was considered minor (level 1) given its limited scale, but signs of potential coordinated amplification of partisan content and inauthentic engagement. In contrast to previous incident 4.2.6, the motive appeared to be fostering a favourable view of the conservative party within the Chinese-Canadian community. These accounts13, all with less than 300 followers, repurposed material from other platforms such as X and TikTok, translated and reposted it with added Chinese subtitles and commentary criticizing the liberal party (see [Screenshots 13\)](#page-53-1). Many of these accounts also left identical or near-identical comments under election-related posts, suggesting formulaic engagement (see [Screenshots 12](#page-52-1)). Much of the translated content included misleading or decontextualized claims targeting the Liberal Party and Carney, which raised concerns about the spread of partisan misinformation, particularly within Chinese-speaking communities. For example, some claims accused Carney of being connected to [Epstein](https://www.bbc.com/news/world-us-canada-48913377) (see [Section 4.2.8](#page-55-3) for the story). Worth noting, the nature of RedNote's algorithm14, which does not privilege follower count, may have further enhanced the reach of such content, even from low-follower accounts. The impact of these posts was relatively minor, with most content receiving only a few likes and limited engagement. Notably, many of the accounts became significantly less active, or stopped posting altogether, once the election concluded, further raising questions about their purpose and authenticity concentrated during the election period.
#### **Screenshots 14.**
**Comment from two different accounts show overlaps, claiming that the Liberal Party profits by bringing in refugees and fake Indian students, and that Carney devised and promoted the carbon tax to benefit his own environmental company, wrecked the UK economy and was pushed out of office, holds three passports, and was a frequent guest of Epstein.**
#### **4.2.8. Surge in faceless accounts and deepfakes of politicians**
**Themes:** Generative AI/deepfakes, disinformation, TikTok
The proliferation of AI-generated political content, ranging from deepfakes of politicians to entire accounts primarily posting synthetic media, was reported during the election. Although most posts appeared satirical in nature, this was considered a minor incident given the volume of these posts and the potential for coordinated activity. The motive appeared to be swaying public opinion and influencing election outcomes, through the tactic of generating and sharing deepfakes of politicians intended (often through faceless accounts). While the origins of this content is vast and unclear, our analysis found that these images were rare during the campaign (one exception was AI-generated images showing Carney with Epstein, see
#### **Screenshots 15.**
**Suspicious account and post (left features a direct chinese translation of content posted by Canada Proud, hinting at Carney's association with Epstein; Bio of the account on the right: "You are welcome to repost and share my content."**
Section [4.2.9\)](#page-56-1). However, deepfakes of politicians used for other purposes were more common — as fake ads and articles linked to cryptocurrency scams (see Section [4.1.1](#page-39-1)), or as memes for satirical purposes. In order to better understand the scope of AI-generated content and the actors behind its dissemination, we collaborated with the [Pol Comm Tech Lab](https://www.polcommtech.com/) in analyzing nearly 700 TikTok accounts that frequently posted about the election but were not established political entities or organizations (more information about the types of accounts included is provided [here](https://www.polcommtech.com/research/influencers-and-elections)). We found that approximately 30% of these accounts were faceless or published primarily AI-generated content, often in the form of memes accompanied by background music or synthetic voice-overs. While many of the faceless accounts appear to be administered by real individuals, their anonymity enables them to share divisive and false content without facing any repercussions or reputational costs. As such,
**Screenshots 16: Examples of AI-generated image sharing where the sharer appears to believe the image is real.**
the content they posted was generally more negative in tone, repetitive in format and themes (e.g., showing high political distrust or attacking party leaders), and more likely to include mis- and disinformation. Overall, while convincing deepfakes were rare and are unlikely to have had a significant impact on public opinion, the proliferation of AI-generated content and faceless accounts can give viewers a sense of a more polluted information environment. We expect deepfakes to become more convincing and harder to detect by the next election cycle, given the release of new AI video generators that are now able to generate voices that match the images just after the election.
#### **4.2.9. High engagement in false claims of Mark Carney-Jeffrey Epstein relationship**
**Themes:** Generative AI/deepfakes, disinformation, bots, X
Between March and April 2025, we observed a surge in suspicious posts on social media spreading false claims that Carney had ties with Epstein (a financier and convicted sex offender) and Maxwell (Epstein's former partner and a convicted sex trafficker), accompanied by [legitimate pictures of Carney](https://www.thestar.com/politics/federal/why-conspiracy-theorists-are-trying-so-hard-to-falsely-link-mark-carney-and-jeffrey-epstein/article_fc78359c-276b-439e-9b1e-fea2858b7ec0.html) [and Maxwell at a music festival in 2013](https://www.thestar.com/politics/federal/why-conspiracy-theorists-are-trying-so-hard-to-falsely-link-mark-carney-and-jeffrey-epstein/article_fc78359c-276b-439e-9b1e-fea2858b7ec0.html) and [AI-generated](https://perma.cc/5JK7-ZEZ9) [images](https://perma.cc/5JK7-ZEZ9). Classified as Level 1, this incident was escalated as we observed deliberate attempts from spam accounts to push the narrative and significant spread (including the spread to offline, explained shortly). The primary motive appears as though it was to damage Carney's reputation and sow doubt among Canadian voters about his connections to Epstein. We traced the origins to understand the tactics being used. Among influential Canadian
users online, the earliest mention appeared on Jan. 6, the day Trudeau announced he was stepping down. However, the story gained momentum in early March, with two accounts posting over 600 identical messages in a single day. Furthermore, we found three inauthentic accounts flagged by [ResetTech](https://thelogic.co/news/canada-election-x-misinformation/) had also contributed to the conversation. While the story itself received over 19 million views on X and at least 100,000 on YouTube, and also spilled offline when a [heckler questioned](https://x.com/RT_com/status/1906024881415680509) [Carney over these links at a rally on March 29](https://x.com/RT_com/status/1906024881415680509), the suspicious accounts themselves received minimal overall engagement, given the this level of engagement and seemingly minimal coordinated efforts, we decided not to escalate and close the incident. Although no official response from X was noted, many posts containing AI-generated images of Carney with Epstein or Maxwell were accompanied by community notes warning viewers of the false images. This event was classified as minor despite reports of similar bot activity coming out from Reset, Tech, and Financial Times. Due to the attention given to this incident, we have provided details of our analysis underpinning our judgement to leave this as a minor incident.
# **SPOTLIGHT**
As part of our post-election analysis, we dug deeper into identifying influence campaigns in promoting the Carney-Epstein association. Our strategy involved analysis of five different behavioural traces: coordination of URL sharing (including quote tweeting), coordination of reply targets, overall account posting volume, single-day posting volume, and user suspensions.
**Figure 27: Timing of unusual activity promoting the Carney-Epstein claims.**
### **Percentage of posts and views per day from identified accounts among all tweets collected**
We found evidence that accounts involved in inauthentic and coordinated activity had spammed copy-paste messages in the days prior to the story gaining traction and promoted the story by replying to popular users on X throughout the entire period. However, we attribute initial traction of the story to popular right-wing influencers and, overall, find posts by suspicious accounts got little engagement by garnering less than 1% of the total views.
According to some of the **[latest methods in uncovering influ](https://dl.acm.org/doi/pdf/10.1145/3589334.3645529)[ence operations](https://dl.acm.org/doi/pdf/10.1145/3589334.3645529)**, accounts involved in influence campaigns typically exhibit similar patterns in URL sharing. To identify coordinated behaviour, we drew network maps where each node was a user and each edge between users represented their similarity in URL-sharing or reply behaviour. We then filtered users to keep only those belonging to densely connected sub-networks.
Overall, we identified 41 users exhibiting high URL-sharing coordination. Collectively, they were responsible for publishing 479 posts and sharing 455 links. We did not, however, uncover a network of users with uniquely similar reply targets. Users in the URL-sharing networks primarily quote-replied (80% of overall posts in our dataset), sharing other popular tweets pushing the conspiracy theory in the reply section of politi-
cians (e.g., Carney or the Liberal Party account), news outlets (e.g., National Post) and popular right-wing influencers (e.g., KatKanada\_TM or govt\_corrupt). The quote-replies tended to add little original content, often repeating the original message. Interestingly, among posts from accounts in the network, less than 1% of them were original posts (i.e., not replies or quotes). The top URLs pushed by the network of users resolved to a tweet claiming **[Carney had been in Epstein's list of contacts](https://x.com/melissalmrogers/status/1904505216344432882)**, and a tweet with **[pictures of Carney and Maxwell at a music](https://x.com/shadowofezra/status/1898880210398826792) [festival in 2013](https://x.com/shadowofezra/status/1898880210398826792)**.
X regularly suspends accounts as part of their effort to rid the platform of users who violate **[X's platform rules](https://help.x.com/en/managing-your-account/suspended-x-accounts#why-might-Twitter-account-suspended)**. This includes spam, abusive, and compromised accounts. As a third party, it is impossible for us to tell why an account was suspended, but for the purposes of this investigation, we made the simplifying assumption that suspended accounts had been involved in inauthentic activity. We identified 163 suspended accounts responsible for 673 posts, of which 25% were from a single user. Again, we found the vast majority of posts from suspended users were replies (44%) or quote-replies (37%), while the rest were quotes. We found suspended users mostly repeated the same message or quoted the same tweet in the reply section of the same set of users as the URL network identified above. This suggests that X identified these as suspicious accounts and eventually removed them.
Users engaged in spam behaviour tend to post far more frequently than ordinary users. We filtered down to users who had an overall daily posting activity of at least 175 posts, replies or retweets per day across their timeline. We call these users overall high-volume posters. We also identified users who had posted over 90 times in a single day on the topic of Carney and Epstein. We call these users topical high-volume posters.
We identified 171 overall high-volume posters responsible for 431 posts, of which, again only a small portion were posts (i.e., not replies or quotes) with the majority being replies, quotes, or quote-replies. Furthermore, we also identified three topical high-volume posters who posted a total of 1,703 times, of which 90% were replies. Alarmingly, we found that almost half of the posts (823) posted before March 10 and 80% of these posted on March 7, were from a single topical high-volume poster who repeatedly posted the same message spreading the conspiracy that Carney had close ties to Epstein.
Suspicious accounts contributed to the conversation by spreading the Carney-Epstein association in the reply section of posts by influential X users by pushing other user's tweets calling into question their ties, or by repeating the same message over and over again. We found that they posted only rarely on their own timelines. We can infer the strategy was to push their message in replies of popular users with the hope of funneling attention to other existing content spreading the allegations. This strategy also overlaps with **[DFRLab's inves](https://dfrlab.org/2025/04/25/bot-like-activity-targets-canadian-political-parties-and-their-leaders-ahead-of-election/)[tigation](https://dfrlab.org/2025/04/25/bot-like-activity-targets-canadian-political-parties-and-their-leaders-ahead-of-election/)** uncovering bot-like activity of users spamming the reply section of politicians and their parties. Below, we present the overall volume of posts per day from suspicious accounts per content type and detection method.
As shown in **[Figure 27](#page-57-0)**, we find that, in total, posts from suspicious accounts accounted for 0.73% of all views garnered by the story on X, with their contribution to the overall posting volume and views across our dataset varying over time.
In line with our initial findings, we find that at the start of March when the overall post volume and view counts were low, posts from a single suspicious topical-high-volume account generated a large portion of the total volume of posts and views surrounding the allegations. While concerning, we note that the story only gained significant traction when influential users posted about the allegations. Given the little engagement that suspicious accounts got, it is likely influential accounts were the main spreaders of the conspiracy. We also note that after the initial days and as the story gained widespread traction, suspicious accounts only garnered less than 1% of the daily views overall.
#### **4.2.10. Inauthentic accounts on X sharing right-leaning election content**
**Theme:** Bots, disinformation, X
Over the course of the election, we flagged multiple anonymous accounts posting a very high volume of Canadian politics content on X that appeared to be inauthentic, i.e., they used AI-generated profile pictures and demonstrated other inauthentic-account characteristics including no profile descriptions, regularly changing profile pictures, and vague bio endings in "No DMs." These accounts were pushing a strong partisan affiliation exclusively anti-liberal content, pro-Trump content and criticizing Carney for not participating in the TVA debate. We classified the incident as Level 1 because this was evidence of inauthentic activity online. However, upon further research, we didn't escalate because these accounts did not receive significant engagement nor showed evidence of coordinated activity. Several of the accounts we had identified were suspended while others remained active and had not been taken down by X early in the election period.
#### **4.2.11. Anti-Carney and Liberal Party networks on X**
**Themes:** Bots, generative AI/deepfakes, misinformation, X
As the election period evolved, we documented and investigated bot activity in tandem with reports from [ResetTech](https://thelogic.co/news/canada-election-x-misinformation/), [DFRLab](https://dfrlab.org/2025/04/25/bot-like-activity-targets-canadian-political-parties-and-their-leaders-ahead-of-election/) and the [Financial Times](https://www.ft.com/content/743c1dec-0c5c-4c39-a940-830ce156c235), and [CAAD](https://caad.info/analysis/briefings/briefing-note-bot-activity-in-canadian-election/) who uncovered networks of accounts and bot-like activity on X promoting anti-Carney content on X. We categorized this as a Level 1 incident as these reports provided evidence of attempts to use inauthentic and coordinated networks of X accounts to manipulate user opinion on X. Furthermore, an [incident](https://caad.info/analysis/briefings/briefing-note-bot-activity-in-canadian-election/) involving bot-like accounts posting "DO NOT ELECT MARK CARNEY" gained enough traction to be trending on X. Our assessment revealed users online were quick to call out the trending tweets and, overall, accounts generally received low-levels of engagement. Combined with findings from other reports suggested this event stemmed from small networks with few accounts. We concluded there was insufficient evidence to suggest these claims were part of a large-scale influence campaign, and that these activities were simply small, isolated incidents.
#### **4.2.12. Liberals planting anti-Conservative buttons and election integrity concerns**
**Themes:** Domestic manipulation, misinformation
Near the beginning of April 2025, Canadian and International news outlets reported two Liberal staffers had planted buttons at a Conservative gathering in Ottawa that were intended to
misrepresent the views of Conservative supporters. This incident, "the [buttongate](https://www.cbc.ca/news/politics/liberal-oppo-csfn-1.7509217) incident," was classified as Level 1 as there was significant engagement with the story and it was an attempt at (offline) information manipulation. We collected approximately 5,000 posts on X in the days after the incident was reported. We found that while most posts expressed negative sentiment, especially about the fact that [the staffers didn't get](https://twitter.com/SeeClickFlash/statuses/1911806097830396283) [fired](https://twitter.com/SeeClickFlash/statuses/1911806097830396283), we didn't find any evidence that the incident triggered any large-scale distrust in the election process. We found that over 80% of the links shared were from trusted news sources and the tweets dominating the conversation were from [cred](https://x.com/natnewswatch/statuses/1911463333833724137)[ible news outlets](https://x.com/natnewswatch/statuses/1911463333833724137) and [journalists](https://twitter.com/BryanPassifiume/statuses/1911569486697668629). Finally, we found that the story garnered a large amount of engagement with posts on X about the story garnering over 5 million views within five days. This engagement appeared organic and we find no evidence of artificial amplification nor linkages to claims of election fraud or the overall election integrity.
# **4.3. Potential incidents (level 0)**
Fourteen cases were identified as potential incidents (Level 0). Level 0 potential incidents are not escalated when our preliminary investigation demonstrates that there is limited discourse, credibility, or engagement online, or that the potential threat identified is illegitimate or unlikely to be consequential for the overall information ecosystem. Of the 14 potential incidents that did not meet the threshold for escalation to minor, moderate, or major incident, we found potential dis- and mis-information campaigns were the most prominent, making up half the cases identified (seven). Other cases included potential foreign influence, extremism or unrest, and domestic manipulation. Most
cases were reported in the latter half of the campaign period, peaking on April 22 (the last week before the election).
Potential dis- and mis-information campaigns were often predicated on misleading or misinterpreted information, revolving around political figures, such as Carney or Poilievre. For example, Carney's vow to address online hate through introducing legislation to investigate potential pathways to online platform regulation was interpreted as a way for him to ban social media and limit free speech. Beyond online hate legislation, other controversial policies led to the amplification of misleading, malicious, or false information. This included Poilievre's proposal for the deportation of foreign visitors who break the law in Canada, which was likened to widespread immigration crackdown occurring in the U.S.
Foreign influence claims were frequently made regarding connections between political candidates and foreign states such as India and China. Various online reports alleged that political candidates, often racialized individuals, may have connections to foreign states or the Khalistan movement. TikTok and Facebook, were also frequently discussed with concern as environments crucial to spreading and amplifying misleading content or facilitating the organization of suspicious online movements. This included reports of using Facebook groups to disseminate false CBC publications to support anti-Liberal Party narratives.
While these Level 0 potential incidents did not escalate further, we continually monitored these events throughout the election period for changes to the nature of these potential threats that would merit escalation.
Bloc Quebecois supporters at an election night rally at the party headquarters in Montreal on April 28, 2025.
# **5. Findings and limitations**
This section expands on the key findings of our analysis identified in our executive summary and reflects on the limitations that should guide their interpretation. In the first part, we present major findings, and in the second part, we turn to the limitations of our approach, recognizing the methodological challenges, data constraints, and interpretive cautions that temper the conclusions.
## **5.1. Findings**
The evidence reveals an electoral information environment that was highly active, uneven across platforms, and shaped by digital and technological disruption but also societal resilience. The findings highlight five overarching themes: the role of influencers, the role of AI, polarization, foreign interference, and information manipulation on social platforms.
#### **5.1.1. Influencers are now the loudest voices in the online political information environment**
Influencers, not parties or newsrooms, dominated distribution and attention across platforms during the campaign. They accounted for approximately almost half (47%) of all political content posted in our dataset, with news outlets at 28% and politicians at 18%. This shift is consistent with the news-sharing ban on Meta and reduced visibility of news on X, which elevated non-news voices and "infotainment" publishers (e.g., Narcity, blogTO). Engagement is also highly concentrated among a small set of accounts. Network analysis shows influencers bridge otherwise separate communities (e.g. political and media groupings), spreading narratives across partisan clusters. This rise in influencers coincides with Canadians being much more active online, but with less trust in the sources and reliability of what they encounter.
### **5.1.2. Generative AI—powered misinformation emerged as a key challenge**
We observed widespread AI use: fake news sites and ads mimicking CBC/CTV; deepfakes of leaders; and faceless TikTok accounts posting AI-generated meme/voice-over content. The majority of the use was for satire, humour, or financial grift, but we also observed multiple instances of AI-generated content intended to mislead Canadians. Canadians noticed this volume, with approximately 80% of reporting concern about AI-generated misleading content (slightly higher than concern about misinformation generally). While the immediate effect on vote choice appears limited, we assess this as a "warning stage": the quality and volume are rising, detection and platform enforcement lags, and the tactic is scalable and easily accessible.
### **5.1.3. Political and platform segregation was evident during the campaign**
Electoral discussion was segmented by platform and ideology: Conservative-aligned voices dominated X, while Liberal-aligned voices partially migrated to Bluesky; TikTok remained marginal for candidates but effective for news outlets and influencers. Our avatar observations revealed X to be the most political and disinformation-dense venue, TikTok with strong algorithmic siloing, Facebook as older echo chambers with economic-nationalist themes, and Instagram as a mobilization hub. This segmentation risks hardening opinion divides and weakening resilience to manipulation.
#### **5.1.4. Foreign interference was a central public concern but interference attempts appeared minor**
We found no evidence of material covert foreign interference affecting outcomes. Observed attempts by foreign actors achieved limited engagement. However, U.S. politics (Trump, tariffs, 51st state rhetoric) featured heavily in Canadian discourse and affected the voting intentions of millions. Survey data showed
that almost half of Canadians (49%, up from 17% in 2021) believe the U.S. interfered in the election, with smaller but notable increases in concern about China (21% to 28%) and Russia (12% to 20%). Online discourse emphasized China and India much more than American interference, however.
### **5.1.5. Targeted manipulation attempts and high volume of claims questioning election integrity are undermining public trust**
In the latter half of the campaign, claims of systemic election fraud surged across platforms. We identified 1,533 posts by influential Canadian accounts making fraud claims, garnering a very conservative 500,000 likes and significant (one in four Canadians) reported exposure. Although most posting occurred on X, engagement was significant on YouTube (47%), X (45%), and also TikTok (6%), reflecting the traction of video formats. Influencers drove almost all ( 90%) fraud-claim posts and most engagement (85%), with the top five (all Conservative favouring) influencers capturing 59% of engagement while political candidates generally did not engage in even speculation on election integrity. Given the volume, we saw a small post-election decline (3%) among Conservative supporters but note that overall trust in election administration remains high.
## **5.2. Limitations**
While the project generated important insights into the dynamics of the election information ecosystem, significant limitations and challenges shaped the scope, depth, and reliability of its findings. These constraints stemmed from both internal research processes and external structural factors. Internally, the rapid pace of the election cycle required trade-offs between speed and rigor, while the influx of new researchers to scale up operations and reliance on experimental methods (such as avatar-based monitoring) introduced unevenness in data collection and analysis. Externally, restrictive platform governance, opaque algorithms, and the profound interconnectedness of the global information environment posed barriers to measurement and interpretation.
These limitations help place the findings in perspective and highlight the need for caution in generalizing conclusions. Notably, foreign interference and misinformation can also occur outside the open online information environment and we have no visibility beyond publicly-available open-source data. They also point to structural issues that will continue to affect election monitoring efforts in Canada: limited data access, fragile methodological infrastructures, and evolving technological and geopolitical threats. Recognizing these challenges
is essential for interpreting the evidence presented here and for improving future cycles of resilience and response-oriented election research.
### **5.2.1. Research, innovation and operational constraints**
Some of the more operational challenges faced by the team included balancing competing priorities, rapid onboarding and innovating in real-time. We faced challenges weighing the need for rigour in research versus the need to share research findings in a timely manner. On the one hand, coalition partners and academic collaborators placed strong emphasis on methodological rigor and analytical depth. On the other hand, the election monitoring environment required rapid turnaround, real-time incident tracking, and immediate insights. This tension often resulted in trade-offs, often with the priority being speed over depth. Adding to this difficulty was the surge of new researchers onboarded during the election period to scale up our ecosystem monitoring capacity. With limited time and resources available for training, many research assistants were thrust directly into projects without extensive preparation. This was especially challenging for the newly formed avatar-based monitoring team, which had to learn both technical and methodological aspects of the work while building realistic online profiles and simultaneously trying to track incidents in real time.
#### **5.2.2. Avatar-based monitoring challenges**
The avatar methodology presented its own set of limitations. While we took significant steps (see [Section 1.2\)](#page-9-1) to build a collection of accounts that were representative of the Canadian public, Avatars' identities and feeds were constantly shaped by platform algorithms, which often shifted them in directions that were difficult to control. For example, X accounts are frequently skewed toward right-wing content, regardless of initial calibration. Avatars were sometimes repurposed to investigate incidents, which inadvertently altered their algorithmic "personalities" and introduced biases. Maintaining stable and authentic virtual identities under such conditions proved extremely difficult.
Recruiting research assistants who could authentically represent Canada's demographic and linguistic diversity was another constraint. Despite efforts, gaps remained. For instance, the inability to recruit a Ukrainian-speaking RA limited monitoring of certain communities. This reduced the breadth of perspectives the avatars could realistically capture, constraining the inclusiveness of the methodology.
Technical hurdles further complicated the avatar strategy. Several accounts, particularly on Meta-owned platforms like
Facebook and Instagram, were flagged, suspended, or removed during the study. Rebuilding these accounts interrupted the algorithmic calibration process and introduced gaps and/or delays in the ongoing monitoring. These barriers highlighted how increasingly restrictive platform governance and researcher accessibility complicate independent, observational research, raising important questions about the sustainability of avatar methodologies in regulated digital environments.
### **5.2.3. Analytical challenges in measuring influence of international voices**
Another limitation was the difficulty of distinguishing Canadian and U.S. influence within online ecosystems. Because Canadian and American information environments are deeply intertwined, organic exchanges blurred the lines between domestic advocacy and international influence and sometimes interference. This entanglement is particularly problematic when U.S.-based actors focus on Canadian issues where they have little understanding, context, or interest in providing factually accurate information. Indeed, a [recent report](https://www.theglobeandmail.com/politics/article-federal-election-tiktok-influencers/) has found TikTok influencers based outside Canada attracted sizable Canadian audiences during the election period while commenting on Canadian political debates. Their content often blended legitimate political critique with mischaracterization of Canadian policy issues. Towards the end of the election period and after the result was called, we have also seen posts about Alberta independence and western separatism from U.S.-based commentators who positioned themselves as advocates for freedom. In both cases, it is difficult to determine whether these narratives should be treated as ordinary cross-border commentary, as a deliberate form of foreign influence, or somewhere in between.
#### **5.2.4. Data constraints**
Access to timely and comprehensive data was the most severe constraint on this project. Data access in Canada is now at its lowest point in at least a decade. Trace data was necessarily limited to a subset of all Canadian voices and engagement measures were delayed by up to three days, undermining the ability to capture incidents in real time or provide rapid responses to emerging threats. Coding dictionaries, while carefully developed, were necessarily non-exhaustive, leaving blind spots in incident detection. Most critically, platforms have rolled back transparency tools, restricted APIs, and imposed governance barriers that have reduced researcher access to a historic low. This degradation of access represents a democratic accountability crisis. Without reliable, timely, and independent visibility into how platforms shape information flows during elections, Canadians are left dependent on opaque processes at precisely the moment when public trust and scrutiny are most needed.
### **5.2.5. Limits of open-source intelligence**
Finally, it is important to emphasize that this project relied exclusively on open-source intelligence. While OSINT provides valuable, timely, and transparent insights into the digital election environment, it offers only a partial perspective. Other forms of investigation, including classified intelligence assessments, law enforcement inquiries, and offline ethnographic research, may reveal dynamics that are not visible through public data, or could even contradict the patterns we identify here. Our findings should therefore be interpreted as one evidence base among many, highlighting observable dynamics in the open online sphere but not providing a comprehensive or final account of foreign interference or information manipulation in the election.
# **6. Recommendations**
The findings of this project make clear that Canada's electoral information environment is increasingly fragile, shaped by rapid technological change, declining platform transparency, and persistent attempts to manipulate public discourse domestically and internationally. While democratic institutions remain resilient, the vulnerabilities exposed during the election highlight the urgent need for proactive and coordinated action. The following recommendations are directed at governments, social media platforms, influencers, civil society, and the general public, with the shared goal of strengthening public trust, building vigilance, ensuring accountability, and safeguarding electoral integrity in the face of evolving threats.
### **Governments must act decisively to restore democratic accountability in digital spaces:**
- 1. Require full disclosure of all political and financially motivated or consequential online advertising at all times, not only during election campaigns. This disclosure should include clear attribution of sources and verified advertiser identities.
- 2. Empower regulators to audit and strengthen platform practices, assess compliance with these requirements, and impose meaningful penalties for repeated failures.
- 3. Ensure that accredited researchers and election authorities have timely access to election-related platform data, so that independent oversight of digital information environments remains possible during critical democratic moments.
- 4. Set clear, enforceable policies to prevent misuse of AI-generated content (including chat bots and synthetic media), requiring transparency (e.g. mandatory
disclosure), accountability, and safeguards against manipulative or deceptive uses in democratic processes. Regulations must flexible and agile to surpass the rate of digital innovation and adoption
**Social media platforms must take far more proactive, collaborative, and transparent steps to protect elections:**
- 1. Dramatically strengthen both automated and manual (human) moderation to detect AI-generated deepfakes and impersonations in real time, particularly those that mimic political figures or legitimate news outlets. Platforms should identify and label all political ads not just those disclosed by the advertiser — especially when funded by foreign actors. Relatedly, platforms should require verified identity checks for all advertisers referencing political content or news organizations, and apply stricter review policies to new advertisers during election periods.
- 2. Provide regular, detailed reports on the volume and type of removed content to enable independent auditing, similar to that [done in the European Union](https://transparency.dsa.ec.europa.eu/statement?lang=fr).
- 3. Enhance and standardize fact-checking functions to ensure timely and accurate verification of content and reduce the spread of misinformation. This includes flagging and labelling false or misleading claims related to the voting process, candidates, and parties.
- 4. Build relationships with election integrity initiatives to ensure accountability and collaboration in addressing these challenges.
**Traditional media must** build more trust with their audiences, and take ownership for their role in shaping the information ecosystem:
- 1. Invest in newsroom training and education to ensure journalists can identify, investigate, and report on information manipulation and information ecosystem dynamics and influences responsibly.
- 2. Provide sustained and neutral coverage of these issues so that Canadians understand how they affect the broader information environment and are better equipped to critically evaluate the content they encounter online.
- 3. Continue holding governments, technology companies, researchers, and influencers accountable on these topics, without fear of retaliation or undue influence.
- 4. Adapt to evolving patterns of trust and news consumption, ensuring that accurate, independent, and unbiased reporting remains accessible and relevant to Canadians.
**Influencers must** embrace their democratic responsibility**.**
- 1. Model responsible digital behavior by demonstrating how to critically evaluate information and encouraging respectful civic dialogue.
- 2. Disclose clearly when content is sponsored or part of a paid partnership, so that audiences can make informed decisions about independent opinion versus advocacy.
- 3. Verify information before amplifying claims about elections, candidates, or public institutions, and refrain from sharing or speculating when accuracy is uncertain. Avoid amplifying outrage-driven or sensationalist narratives that may originate from coordinated disinformation campaigns.
- 4. Use their reach to elevate credible information on voting, democratic participation, and civic education, ensuring audiences have access to timely and accurate resources.
**Civil society and researchers must** strengthen their role in safeguarding elections.
- 1. Engage in anticipatory monitoring and prevention through mapping likely misinformation narratives, such as claims of fraud, foreign interference, or AI-driven manipulation, with corrective materials prepared ahead of time.
- 2. Expand civic education efforts well beyond formal schooling to include digital literacy and misinformation awareness in community centers, workplaces, and among vulnerable populations.
- 3. Deepen collaboration between researchers, journalists, creators, election administrators, and community leaders to coordinate responses, share best practices, and reduce duplication of effort. By working collectively, they can build resilience against the fast-evolving threats that will shape future electoral cycles.
**The Canadian public** must seize their role in safeguarding democratic discourse.
- 1. Broaden their information environment through seeking out diverse perspectives — including those that challenge their assumptions — to counteract filter bubbles, reduce bias in how they process information, and strengthen their capacity to evaluate competing claims. Build awareness of how algorithms, platforms, and influential voices curate and amplify content while recognizing one's own role in shaping what you see and believe.
- 2. Practice informed vigilance by considering the credibility and motives of sources, especially when content seems engineered to provoke outrage or fear, and staying alert to coordinated disinformation campaigns.
- 3. Foster resilience by engaging in respectful offline conversations across partisan and community lines, helping to reduce division, strengthen empathy, and reinforce healthier democratic habits.
**Taken together, these recommendations point to a single conclusion: safeguarding Canada's electoral integrity now requires urgent, coordinated, and sustained action. The information environment has grown too volatile, too opaque, and too easily manipulated to leave democratic resilience to chance.Implementing these measures is no longer optional if Canada is to maintain trust in its elections and preserve the foundations of democratic accountability.**
# **7. Methodology**
# **7.1. Incident response methodology**
One of the Media Ecosystem Observatory's key functions is to monitor, detect and respond to threats to the Canadian information ecosystem. During the election, the Observatory team used a modified approach to our [incident response protocol](https://www.cdmrn.ca/pier). At a high level, the Observatory team (in collaboration with our [Coalition on Information Ecosystem Resilience\)](https://www.cdmrn.ca/coalition-for-information-ecosystem-resilience) monitored the ecosystem to detect harmful information or information activities that could impact integrity of the electoral process (threats). If and when a threat was significant enough, it was deemed to be an information incident, i.e., *a disruption in the information ecosystem, including both sudden and prolonged interruptions, that significantly impacts the normal flow and/or integrity of information, leading to potential or actual harm to the public, government, Canadian democracy, and/or the broader information ecosystem.* Once an incident was declared, in collaboration with the broader Coalition, MEO responded by conducting rapid investigations into the cause, nature and impact of these events. All findings were shared as they emerged, within a two-to-four day window.
For a more detailed understanding of the process, the remainder of this section outlines the process we used in five stages (Monitor and Detect, Declare, Activate, Investigate, Debrief).
1. **Monitor and Detect:** Conducting ongoing observation of the information ecosystem to identify information threats (e.g., contentious topics, emerging trends, threatening entities online, generated content, bots). To do so, a distributed situation awareness approach is used, i.e., a blend of mixed ecosystem monitoring methods including automated and manual, but also knowledge sharing between a strong network of stakeholders established specifically for elections-monitoring efforts, in-hand with technical experts.
- E While monitoring the information ecosystem for near real-time detection can be challenging, we conduct observations from various data sources to leverage the most recent and accurate information as possible. This includes crowdsourced reporting from our research networks and election coalition partners, establishing a public information hotline for the general public to report emerging threats they detect, ongoing manual and automated data collection and analysis from various social media platforms, and assessing trends through tracking surveys.
- E **Situation Reports:** The incident response team develops a daily situation report (SITREP), which is broadcasted to all internal members of the team at the beginning of each day. The SITREP provides updates on ongoing and new potential incidents and their incident level — noting if there is new pertinent information or changes to the threat level as identified through our investigations. It also identifies the originating source of the tip or information (e.g., Tipline, avatar-based social media monitoring team) and lists the key contacts and individuals responsible for overseeing the investigation or assessment of each of the potential incidents. It provides a brief overview of key changes to the media ecosystem and a preliminary analysis of any trends that may be emerging. The purpose of the SITREP is to provide a systematic and timely daily update to ensure that all team members are apprised of the Observatory's activities.
- 2. **Declare**: Evaluating and triaging threats as potential and actual incidents (based on a [clear set of criteria](https://ojs.iscram.org/index.php/Proceedings/article/view/193/167)), and evaluating the response requirement and capability to declare an incident response.
- E A potential incident is assessed against four criteria (spread, incident complexity, potential harm and impact, and response requirement, and complexity)
| Criterion | Level 1 (Minor) | Level 2 (Moderate) | Level 3 (Major) |
|------------------------------|--------------------------------------------------------------------|-------------------------------------------------------------------------------------------------------------------------------------|--------------------------------------------------------------------------------------------------------------|
| Spread | Slow spread and activity
localized or niche
communities only | Medium spread with noticeable
engagement and activity across one
or two platforms with some media
and influencer attention | Rapid or viral spread and activity
across multiple platforms with major
media and influencer attention |
| Incident complexity | A single simplistic
misleading claim or
manipulation | Moderately deceptive claims or
manipulations that may include
partial truths mixed with falsehoods | A sophisticated and coordinated
overlapping set of claims or
manipulations |
| Potential harm
and impact | Minor risk to any group or
the public at large | Could cause moderate public con
cern or confusion, with potentially
significant risk to specific groups | Could undermine election integrity,
incite violence, or severely destabilize
public trust |
Requires a multi-disciplinary
team to address
**Table 4. Decision making matrix for incident declaration and triage (from [Phillips & Chan, 2025](https://ojs.iscram.org/index.php/Proceedings/article/view/193/167)).**
to evaluate a) if a threat meets the threshold for an incident, and, if so, b) the severity of that incident.
Contextualization and clarity are sufficient to address
E Incident severity is triaged as Level 0 to Level 3 (potential, minor, moderate and major incidents):
**Response requirement and complexity**
- i. Potential incidents (Level 0) are events (or threats) that have been flagged for demonstrating characteristics of a threat or vulnerability, but insufficient evidence exists to suggest a significant impact on the information ecosystem. Level 0s are typically noted, archived, and closed, but sometimes are kept open for ongoing monitoring.
- ii. Minor incidents (Level 1) are flagged but the incident response team is not activated; incident reporting consists of sharing more information (including learning content) about the incident
- iii. Moderate to major incidents (Levels 2–3) trigger the need for activation of the information incident response process, a small or larger investigation occurs on a short-timeline depending on the incident
- E Although an incident may be flagged as moderate or major, a formal activation of incident response occurs when the response requirement and capability are met ([Table 5\)](#page-67-1).
- E For election monitoring purposes, we re-evaluate incident levels for every identified potential incident every 24 hours based on emerging information.
- 1. **Activate**: Initiating response through activation and coordination of an incident response team (interdisciplinary group of subject matter experts) to design and execute a rapid research investigation, scaling up data collection and issuing a public notification.
- E Once an incident is declared and response capability is deemed to be sufficient, an activation message is sent out to coalition partners to request immediate participation in the incident investigation (within 24 hours)
Requires immediate large-scale mobilization of a wide team
- E The incident response team gathers specialized coalition members, subject matter experts, and technical specialists to support an incident response. A team specific for each incident is formed. An investigation is planned, including the identification of research questions, delegation of roles and resources, and outlining timelines. Bespoke data collection may be required as part of this process.
- E Once a declaration of an incident has been established, a public notification is also shared through our information channels. Public updates are provided as the investigation evolves.
- 2. **Investigate**: Conducting and sharing updates to shed insight on the nature and potential impact of the incident.
- E Incident response team members conduct rapid research and deliver their findings as they emerge as incident updates. Once an update is prepared, it
**Table 5. Decision making matrix for response capability.**
| Criterion | Description | |
|-------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|--|
| Reputational, legal,
and infrastructure risk | Responding to an incident can draw scrutiny, potentially leading to unwanted and
malicious attention. At the same time, downplaying a credible threat may erode trust
and diminish credibility. We aim to be transparent and clearly communicate the
decision-making process | |
| Researcher safety | Investigating sensitive or high-profile campaigns may expose individual researchers to
harassment, retaliation, or lawsuits. We consider physical security, digital protection, and
other safety dimensions as we enter into complex investigations, particularly those
involving influential actors or transnational entities | |
| Resource availability | Effective responses demand skilled personnel, analytical tools, and time. Initiating an
investigation with insufficient staff and technology risks incomplete analysis or
delayed intervention | |
| Need for positive impact | A response can reduce harm and increase transparency and contextual understanding.
By providing clarity and evidence-based information, we aim to bolster public trust and
resilience against future manipulation. However, the risk of inadvertently amplifying the
manipulation must also be weighed carefully | |
is shared with the wider incident response team for rapid peer-review and edit. It is subsequently shared with the public. The release of investigation findings begins within 48 to 72 hours of activation (depending on the incident).
- 3. **Debrief**: Synthesising findings into a post-investigation debrief with a birds-eye view summary of the nature of the incident (e.g., origin, tactics, motives, attribution if possible), impact (e.g., scope and reach), and lessons learned.
- E The decision to conclude an investigation is made on a case by case basis, on factors including a drop in overall engagement with an incident, the perceived impact decreases, or the nature of the incident ceases to evolve in ways that warrant further assessment (e.g., an event naturally comes to a close or does not evolve further beyond what has been investigated).
- E A debrief report is drafted for wide-scale circulation, conducting an overall analysis of the incident updates and investigation and synthesising overall findings. The report highlights the main findings: the nature, cause, and impact of the incident; and provides lessons learned.
## **7.2. Survey methodology**
Our team fielded surveys designed to measure Canadians' attitudes and behaviours leading up to, during, and after the 2025 federal election, with a particular focus on the online information environment and related risks such as misinformation and foreign interference. The CDMRN conducted monthly tracking surveys with nationally representative samples of 1,400-1,500 Canadians for over a year prior to the election. These surveys provide a baseline for understanding longer-term trends in media consumption, political engagement, and online behaviour.
The pre-election rolling daily survey was fielded from March 23, 2025 (the day the election writ was issued) to April 27, 2025 (the day before election day), targeting 200 respondents per day. In collaboration with the commercial survey firm Leger Opinion, we obtained a nationally representative sample using quotas based on Statistics Canada benchmarks for age, gender, and region. Post-stratification weights were applied to ensure national representativeness. The final pre-election sample used for analysis consisted of 6,955 responses. The questionnaire covered a range of topics including demographics, social media use, social and political trust, political participation, misinformation, and perceptions of election integrity and foreign inter-
**Table 6. Sample characteristics (unweighted).**
| Category | Target
(%) | Pre-election
(%) | Post-election
(%) |
|----------|---------------|---------------------|----------------------|
| Gender | | | |
| Male | 50 | 49 | 54 |
| Female | 50 | 50 | 45 |
| Age | | | |
| 18–29 | 20 | 19 | 15 |
| 30–44 | 25 | 25 | 28 |
| 45–59 | 25 | 26 | 30 |
| 60+ | 30 | 30 | 28 |
| Region | | | |
| Atlantic | 7 | 6 | 7 |
| Quebec | 22 | 22 | 22 |
| Ontario | 39 | 40 | 39 |
| Prairies | 19 | 15 | 15 |
| B.C. | 13 | 13 | 14 |
ference. Additional question blocks were included for several days following the Leaders' Debate, along with questions on opinions about the CBC.
The post-election survey was conducted from May 12–14, 2025, two weeks after election day. We aimed to recontact approximately 2,000 of the original 6,955 pre-election respondents and successfully resurveyed 2,078 of them. This wave included questions on voter turnout, satisfaction with the functioning of democracy, social media use, awareness of campaign incidents, and political engagement.
Information about the characteristics of the pre- and post-election samples are provided in [Table 5](#page-67-1). All analyses in this report are weighted by age, gender, and region.
In some of the analyses throughout the report, we compare Canadians' responses to those collected during the 2021 federal election. The 2021 election survey, conducted by Dynata, included 7,322 respondents in the pre-election period and 2,814 recontacts in the post-election period.
### **7.3. Digital Trace and social media analysis**
Our [data infrastructure](https://arxiv.org/abs/2506.10942) is designed as a modular and scalable infrastructure for collecting, storing, analyzing, and visualizing digital trace data from multiple social media platforms. The system supports both almost real-time (d-3) and historical data processing and is intended to provide flexible, cross-platform insights into the Canadian information ecosystem. The data pipeline utilizes the seedlist to define and control what data to collect and media crawlers to gather data from different platforms
#### **7.3.1. Seed list**
Our dataset consists of posts collected on X, Instagram, TikTok, Bluesky and YouTube from relevant actors in the Canadian media ecosystem. We refer to these actors as seeds, which we define as a person, group, organization, or media entity that is of substantive interest to Canada's media ecosystem. To be included in the seed list, these entities must be alive, in the case of a person, or in operation at the time of inclusion, for the remaining types. In addition, to remain included in the seed list, the entity must have had activity on at least one platform in the last 12 months.
To identify politicians, we used official rosters of elected members of the House of Commons and all provincial legislative assemblies, collecting each politician's name, province, and political party. Provincial parties were mapped to their federal counterparts to facilitate cross-level comparisons. For the federal election, we used the official list of registered candidates from Elections Canada and collected each candidate's name, province, political party, and riding.
For the identification of news outlets, we used [Media Cloud](https://www.mediacloud.org/)'s "national" and "state-local" datasets of Canadian news outlets. Media Cloud's lists were manually reviewed to confirm that they were still in operation and based in Canada. We attempted to identify as many political influencers who met the following criteria: at least 10,000 followers in TikTok/X or at least 5,000 followers in Facebook/Instagram/YouTube/Bluesky, at least a 1:2 following-to-followers threshold, and have a majority of the content be political. For government organizations, we consulted the official website of the Government of Canada for the list of federal departments and agencies. For civil society organizations, we consulted Revenue Canada's charities list and the Government of Canada's list of unions and filtered only organizations that met the follower threshold or revenue and membership thresholds.
#### **7.3.2. Media crawlers**
To build a comprehensive dataset, we developed platform-specific tools and workflows for collecting data from social media platforms. Metadata for each post was preserved in its entirety and stored in JSON format to ensure structural consistency and completeness.
**X:** We developed a data collection library for X that programmatically queries the platform's advanced search and timeline endpoints to retrieve recent posts. This approach allowed us to collect data on a bi-weekly basis, as well as historical posts from the months prior to an account's inclusion in our seed list, by specifying custom date ranges.
**TikTok:** We used a data collection library for TikTok as described in [Steel et al. 2024](https://arxiv.org/abs/2301.08305), that allows us to do full reverse-chronological collections for each account. The scraper was run at regular weekly intervals to ensure comprehensive coverage of posts while maintaining the integrity of engagement metrics. While an official [TikTok Research API](https://developers.tiktok.com/doc/research-api-codebook) does exist, it was not available for Canadian researchers as of the 2025 election.
**Instagram:** Until Aug. 14, 2024, Meta provided access to its [CrowdTangle](https://about.fb.com/news/2023/11/new-tools-to-support-independent-research/) platform, which we used to collect Instagram data from Jan. 1, 2022, to Aug. 10, 2024. During this period, data collection followed a daily scraping schedule. After CrowdTangle was discontinued on Aug. 14, 2024, we transitioned to a [custom](https://nostarch.com/social-media-observatory) [scraping solution](https://nostarch.com/social-media-observatory).
**Bluesky:** For Bluesky, we utilized the [official API](https://docs.bsky.app/) to collect public posts in a structured and efficient manner. The platform's open protocol and clear documentation allow for streamlined integration into our crawler system. As our newest platform, we currently only collect influencer and politician data.
**YouTube:** We used the [YouTube Data API](https://developers.google.com/youtube/v3) to collect metadata associated with videos posted by accounts in our seed list. This included post-level details such as titles, descriptions, publication timestamps, and engagement metrics like views and likes. To ensure temporal continuity and capture engagement activity close to the time of posting, we implemented a frequent and regular collection schedule.
#### **7.3.3. Topic modelling**
Topic modelling is an unsupervised text‐mining technique that infers latent semantic patterns from large, unlabelled corpora. Rather than requiring predefined categories, the algorithm assumes that each document is generated by a mixture of hidden topics and that every topic is characterised by a distinct distribution over the vocabulary. Classical approaches such as Latent Dirichlet Allocation (LDA) estimate these distributions with
probabilistic inference, but more recent methods replace the bag-of-words representations with dense sentence or document embeddings obtained from large language models. Operating in a continuous vector space makes it possible to capture subtle syntactic and semantic nuances, to work effectively across different languages, and to integrate modern clustering and dimensionality-reduction algorithms that were not available when the original probabilistic frameworks were proposed.
The present study adopts BERTopic, a modular pipeline that combines transformer-based embeddings with non-parametric clustering to yield coherent, human-interpretable topics. In the first stage, each text instance is converted into a high-dimensional embedding using the multilingual variant of MiniLM fine-tuned for paraphrase similarity. These embeddings are then projected to a low-dimensional manifold with UMAP, which preserves the local neighbourhood structure while substantially reducing computational complexity for subsequent steps. Topic discovery is performed with HDBSCAN, a density-based algorithm that automatically determines the number of clusters and isolates noise points; its minimum cluster size parameter is scaled to corpus size to balance granularity and stability.
Once clusters are identified, BERTopic constructs a sparse class-based TF-IDF (c-TF-IDF) representation that contrasts the vocabulary of each cluster with the entire corpus, thereby highlighting discriminative terms even in highly imbalanced data. To enhance interpretability, the framework ensembles multiple representation strategies: A KeyBERT-inspired extraction ranks candidate keywords by their cosine similarity to the centroid embedding, while a Maximal Marginal Relevance step re-orders them to maximise semantic coverage and minimise redundancy. The top keywords form concise, descriptive labels that are appended to the numerical topic identifier, and each document inherits the label of its most probable cluster together with an assignment confidence.
### **7.3.4. Election fraud and foreign influence claims identification**
Seedlist posts containing foreign interference and election fraud were identified using a two-step process. Seedlist posts between February 23 and May 28, 2025 (excluding those from foreign sources as well as sports) were initially filtered using keywords to identify candidate posts related to each topic. These posts were then passed through the large language model, Claude, to verify they were related to the federal Canadian election and were related to foreign interference or election fraud.
The keywords used for foreign interference are given in [Table](#page-71-1) [8](#page-71-1) on page 68. Specifically, a post needed to contain at least one target word as well as at least one context word within a 25-word window around the target word. These words were intended to flag general interference claims. The keywords in Table 9 were further used to flag posts of specific narratives, namely the Carney/China, Tay/Chiang, and Poilievre/India narratives from [section 5.1](#page-60-2). For these posts, a post needed to contain both words in a pair within a 25-word window. [Table 9](#page-72-0) gives the keywords used to identify election fraud. In general, a post needed to contain an election pattern as well as an additional pattern to be flagged.
After identifying candidate posts, Claude was used to verify true mentions of foreign interference or election fraud from any false positives identified by the keyword search. For foreign interference, posts only needed to discuss interference without explicitly claiming it was happening, however for election
fraud, posts were classified into 3 classes: ones that explicitly claimed fraud or cast doubt about the election results, ones that discussed fraud but did not take a clear stance on whether it was happening, and those that did not discuss fraud.
In addition to the classification of the posts, for foreign interference, Claude was tasked with identifying who the author claimed was responsible for the interference. This was not limited to a set of actors but was intentionally left open-ended for Claude to identify any actor. Manual assessments of 50 posts for both election fraud and foreign interference deemed Claude proficient at these tasks.
Although these methods were not exhaustive, our keywords were chosen, wherever possible, to be as broad as possible and thereby minimize false negatives.
**Table 7. Keywords for identifying general interference claims.**
| | English | French |
|---------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Target words | "election", "vote", "voting", "ballot", "polls", "election day",
"canada", "canadian", "elections canada", "canadian
election" | "élection", "élections", "vote", "voter", "scrutin", "urne",
"jour d'élection", "canada", "canadien", "canadienne",
"canadiens", "canadiennes", "élections canada", "élec
tion canadienne" |
| Context words | # Interference-related verbs | # Interference-related verbs |
| | "interfere", "interfering", "interfered", "meddle", "med
dling", "meddled", "tamper", "tampering", "tampered",
"manipulate", "manipulating", "manipulated", "rig",
"rigging", "rigged", "hack", "hacking", "hacked", "com
promise", "compromising", "compromised", "infiltrate",
"infiltrating", "infiltrated", "spy", "spying", "spied", "leak",
"leaking", "leaked"
# Influence operations
"disinform", "disinforming", "disinformed", "misinform",
"misinforming", "misinformed", "propagandize", "propa
gandizing", "propagandized" | "interférer", "interfère", "interféraient", "interférence",
"s'ingérer", "ingérence", "se mêler", "se mêle", "sont
mêlés", "mêlage", "truquer", "truqué", "truquant",
"truquage", "manipuler", "manipulé", "manipulant",
"manipulation", "pirater", "piraté", "piratage", "piratant",
"cyberattaque", "cyberattaques", "compromettre",
"compromis", "compromettant", "infiltrer", "infiltré", "in
filtrant", "infiltration", "espionner", "espionné", "espion
nant", "espionnage", "fuiter", "fuite", "fuitant", "fuites"
# Disinfo & manipulation
"désinformer", "désinformation", "malinformer", "malin |
| | # General suspicion/intelligence | formation", "propagande", "propagandiste", "diffuser de
la propagande" |
| | "influence", "influencing", "influenced", "surveil", "surveil
ling", "surveilled", "monitor", "monitoring", "monitored", | # Surveillance & influence |
| | "monitor", "monitoring", "monitored" | "influencer", "influencé", "influencent", "influence", "sur |
| | # Actors and tools | veiller", "surveillance", "surveillé" |
| | "propaganda", "bots", "troll", "trolling", "troll farm", "espi
onage", "cyberattack", "fake news", "phishing" | # Actors and tools
"bot", "bots", "faux comptes", "troll", "trolls", "ferme
à trolls", "espionnage", "phishing', "hameçonnage",
"fausses nouvelles", "infox" |
**Table 8. Keywords for identifying specific foreign interference narratives.**
| | English | French |
|------------|---------------------------------------------|--------------------------------------------|
| Word pairs | # Carney/China | # Carney/China |
| | "Carney", "China" | "Carney", "Chine" |
| | "Carney", "Beijing" | "Carney", "Pékin" |
| | "Carney", "Chinese" | "Carney", "Chinois" |
| | "Carney", "WeChat" | "Carney", "WeChat" |
| | "Cantonese", "Carney" | "Cantonais", "Carney" |
| | # Tay/Chaing | # Tay/Chaing |
| | "Tay", "China" | "Tay", "Chine" |
| | "Chiang", "China" | "Chiang", "Chine" |
| | "Tay", "Chinese" | "Tay", "Chinois" |
| | "Chiang", "Chinese" | "Chiang", "Chinois" |
| | "Tay", "Beijing" | "Tay", "Pékin" |
| | "Chiang", "Beijing" | "Chiang", "Pékin" |
| | # Poilievre/India | # Poilievre/India |
| | "Poilievre", "India"
"Poilievre", "Modi" | "Poilievre", "Inde"
"Poilievre", "Modi" |
| | "Poilievre", "Indian" | "Poilievre", "Indien" |
## **7.4. Avatar-based social media monitoring**
To complement the holistic and automated social media analysis, we employed an avatar-based social media monitoring method to capture more contextual, real-time and qualitative insight into the online information environment during the Canadian federal elections. While digital trace methods enabled broad trend detection and large-scale mapping of narratives, the avatar approach provided a more grounded view of how platform algorithms and information flows vary depending on who the user is. These two methods were designed to work in a complementary manner: Insights from digital trace analysis helped guide what to look for, while findings from avatar-based monitoring helped surface narratives or incidents that might not yet be prominent at scale.
### **7.4.1. Approach**
Specifically, we created 39 social media profiles, referred to as avatars, designed to reflect the different profiles of the Canadian population in terms of age, gender, region, language, political orientation, ethnic and cultural background and topic interests. Our "on the ground" Boots team used these accounts to actively monitor election-related content across 10 social media platforms, including Facebook, Instagram, X, TikTok, YouTube, Bluesky, Wechat, Reddit, Truth Social and RedNote. Overall, 89 social media accounts were created and used for avatar-based social media monitoring.
#### **7.4.2. Needs assessment**
The design of the avatar cohort was informed by the results of our internal environmental scan (described in Section 1.2) conducted prior to the election. This comprehensive scan assessed the Canadian information ecosystem for emerging threats and vulnerabilities, highlighting key risks associated with election security, foreign interference and most relevantly, the potential targeting of diaspora communities through disinformation and interference campaigns. Our avatar framework was developed based on emergent themes identified in this scan to best reflect the diverse demographic and interests of the Canadian population, specifically considering the representation of communities identified as particularly vulnerable to influence operations.
#### **7.4.3. Avatar creation and summary**
Avatars were deliberately constructed to reflect both the demographic diversity of Canada and the information threats and vulnerabilities observed in prior elections. Overall, we developed 39 unique avatars spanning a range of age groups, regions, political leanings, and ethnic, religious, and cultural identities, several chosen based on their perceived susceptibility to manipulation, including recent immigrants, ethnic minorities, politically disengaged voters and single-issue voters. Because platform algorithms personalize content based on user behavior and identity signals, our avatars required sustained, hands-on engagement across different platforms to properly configure their algorithmic environment. This not only ensured that each avatar's feed reflected what a real user might see, but also allowed us to observe how algorithmic behaviour, and the types of political content surfaced, varied across platforms and user profiles.
- E **Gender representation:** The avatar pool included a diverse gender composition, with 21 avatars identifying as male, 16 as female, and two as other. The gender representation mirrored that of the general Canadian population according to [Statistics Canada](https://www150.statcan.gc.ca/n1/pub/12-581-x/2022001/sec6-eng.htm#) and ensured that gender-based content targeting can be observed and analyzed across different online ecosystems.
- E **Age representation:** The avatar pool included a broad age range, with representation among young adults (nine ava-
#### **Table 9. Keywords for identifying election fraud claims.**
#### **English French # election\_pattern** "vote", "voting", "election", "reelection", "re-election" **# pen\_eng\_pattern** "pen", "pens", "pencil", "pencils" **# disenfranchise\_pattern** "not register", "disenfranchise", "not allowed to vote", "prevent from vote", "prevented from vote", "preventing from vote", "prevents from vote","crossed off", "crossed out" **# voter\_eligibility\_pattern** "not eligible", "non-citizen", "not a citizen", "non-canadian", "not canadian", "voter id", "photo id", "pill bottle" **# tbtr\_pattern** "too big to rig", "toobigtorig" **# ballot\_pattern** "ballot" **# tampering\_pattern** "unseal", "home", "unsecure", "tamper", "stuffed", "stuffing" **# poll\_pattern** "poll" **# fake\_doubt\_pattern** "fake", "inaccurate", "do you believe", "don't believe", "do not believe", "misleading" **# poilievre\_pattern** "carleton", "poilievre", "fanjoy", "battle river crowfoot", "battle river-crowfoot" **# long\_ballot\_pattern** "long ballot", "longest ballot" **# recount\_pattern** "recount" **# election\_pattern** "voté", "élection", "réélection", "ré-élection" **# pen\_fr\_pattern** "stylo", "crayon" **# disenfranchise\_pattern** "privé de droit de vote", "privée de droit de vote", "priver de droit de vote", "privés de droit de vote" **# voter\_eligibility\_pattern** "non-citoyen", "pas canadien' **# tbtr\_pattern** # (English only — omitted) **# ballot\_pattern** "urne" **# tampering\_pattern** "maison", "non sécurisé", "manipulation", "bourrage", "trafiqué", "falsifié" **# poll\_pattern** 'sondage" **# fake\_doubt\_pattern** "faux", "truqu", "inexact", "crois pas", "croit pas", "croient pas", "croyons pas", "croyez pas", "croyais pas", "croyait pas", "croyaient pas" **# long\_ballot\_pattern** "bulletin de vote" **# terrebonne\_pattern** "terrebonne" **# recount\_pattern** "recomptage"
**# fraud\_pattern**
tars aged 18–24) and mid-career adults (24 between 25–55, 13 of which were aged between 45 and 64). Seniors were also included, with 2 avatars aged 65+. This range supported analysis of how political content and platform dynamics vary by generation.
E **Regional representations:** Avatars were distributed across all major regions of Canada. Ontario was the most represented with 11 avatars, followed by British Columbia (BC) with nine, and the Prairies (including Alberta, Saskatchewan, and Manitoba) with five. Quebec had six avatars, and the Maritimes were represented by five avatars.
"volée", "truquée", "manipulation", "manipulée", "falsifiés"
"fraud", "rigg", "cheat", irregularity", "stole", "steal"
**# fraud\_pattern**
- E **Occupation and economic background:** The avatars were designed to represent a broad occupational spectrum, including students, service and blue-collar workers (examples include truck driver, security guard and supermarket cashier), white collar professionals (examples including lawyer, insurance agent and financial analyst), homeworkers and retirees.
- E **Political orientation:** Avatars were created to represent a range of political views, from strongly left-leaning to far right-leaning, as well as more centrist or disengaged positions. Broadly, 11 avatars were right-leaning, eight centrist, and 11 left-leaning. An additional four avatars were politically disengaged, and two represented Quebec nationalist perspectives. This ideological spread helped us assess how narratives, especially polarized or misleading ones, are tailored to political identity.
- E **Key topics of interest:** The most frequently cited issues among avatars include the economy (seven avatars), inflation (six), housing (four), health care, climate change, energy, Canadian sovereignty, and the environment (each mentioned by three avatars). Others engaged with more specific concerns, such as Canada-China relations or Indigenous rights, allowing us to monitor how issue-based content is prioritized or distorted on different platforms.
- E **Ethnic and cultural backgrounds:** To reflect Canada's [di](https://www150.statcan.gc.ca/n1/daily-quotidien/221026/dq221026b-eng.htm)[verse population makeup](https://www150.statcan.gc.ca/n1/daily-quotidien/221026/dq221026b-eng.htm) and identify potential points of targeted or culturally-specific manipulation, avatars were assigned a variety of racial and ethnic identities. These include White (19 avatars), Mixed (three), Indian (two), Chinese (two), Black (two), Ukrainian (one), Russian (one), Jewish (one), Middle Eastern (one), and Chinese from Hong Kong (one). Several avatars were designed as recent immigrants or diaspora community members, helping track how foreign policy, diaspora tensions, or identity politics influence the digital information environment.
**Platform assignment:** Each avatar was assigned to at least two social media platforms to allow for direct comparison of platform-specific content and algorithmic behaviour. Platform selection was guided by the demographic and ideological attributes of each avatar. For example, Canadians above 40 were more likely to use Facebook and YouTube, while younger users were assigned TikTok and Instagram. Ideological leanings also informed platform placement: Left-leaning avatars were placed on Bluesky, while right-leaning profiles were assigned to X, Truth Social, or YouTube, where conservative political content, including populist and anti-establish narratives, is more commonly encountered. Linguistic and cultural background further informed platform placement. For example, Chinese-speaking Canadians tend to be more active on social media platforms such as Wechat and RedNot. Similarly, avatars designed to reflect members of diaspora communities were integrated into language-specific pages, group chats, and channels that mirror the kinds of online spaces commonly used within those communities. These groups often serve as primary sources of information and discussion for their members, and can also be pathways through which culturally tailored narratives, both accurate and misleading, circulate. Together, this strategy enables a nuanced, cross-platform analysis of how digital environments filter political information in relation to user identity.
#### **7.4.4. Avatar-based social media monitoring team**
The team consisted of 12 members, with 11 research assistants and one project lead. The project lead oversaw the design of the avatar framework, provided ongoing methodological guidance and coordinated with the broader Observatory team. The on-the-ground research was carried out by a team of 11 RAs, primarily undergraduate and graduate students from McGill University and the University of British Columbia. One RA served as coordinator, managing internal communications and ensuring daily operations ran smoothly.
The RAs were recruited and hired in early March, before the official election call. All RAs were selected based on their knowledge and academic background in Canadian politics, public policy or media studies, and many had completed relevant coursework on media and democracy. We prioritized candidates with language skills critical to the project, such as Mandarin, Punjabi and Russian, to better reflect Canada's linguistic diversity online. In addition, individuals were hired for their familiarity with different communities, enhancing the authenticity of avatar engagement within culturally specific online space. Each RA was assigned a dedicated phone to be exclusively used to manage three to five avatars across multiple social media platforms.
#### **7.4.5. Account creation**
The initial process of avatar creation took roughly two weeks. This involved carefully configuring the platform algorithm to reflect each avatar's profile, researching and following relevant pages, influencers, and interest-based content to simulate the kind of feed a real person with those traits would encounter. As the election period progressed, we made adjustments to the profiles to ensure they continued to reflect realistic user behaviour. For instance, engaging with (i.e., like, comment,
or repost) non-political content that is related to an avatar's custom, personal interests and creative pursuits that vary from person to person, like ice-hockey memes or fashion trends. Some avatars were delegated to specific platforms, such as Wechat and Telegram, based on particular democratic usage patterns — preferences which can differ based on a group's **Capture and Documentation:** Once identified, relevant content was documented through two main tools: Story Chaser, an internal database for initial content capture, and the Incident Log, a shared workspace for deeper investigation. Story Chaser enabled the team to capture links, text, and visual content encountered on social media platforms, especially in cases where
### **7.4.6. Data collection**
ethnic or regional background.
Our data collection process followed a two-step structure. First, using the set of avatars, the social media monitoring team detected and identified potential information threats and reported them to the broader team. Secondly, we captured and documented content using our internal tools Story Chaser and Incident Log.
**Detection and Identification:** The Boots RAs monitored the social media feeds of their assigned avatars daily to identify narratives, events and accounts that could potentially pose an informational threat to the Canadian online environment during the election period. This includes both organic content (e.g., viral posts, trending hashtags) and potentially coordinated activity (e.g., suspected misinformation campaigns, foreign influence, or bot amplification). The team followed specific internal guidelines to determine whether the content should be flagged. Key questions include: Is the event potentially or actually disruptive to the information ecosystem? As it relates to this election, does this event pose a risk to the electoral process? If the content meets the threshold, it will be flagged for escalation or further review.
posts were later deleted or altered.
### **7.5. Story Chaser data collection and analysis tool**
the Story Chaser tool was developed to enable structured, scalable monitoring of social media and media content related to high-impact or politically salient events. Built using [KoboTool](https://www.kobotoolbox.org/)[box](https://www.kobotoolbox.org/) and the XLSForm standard, the tool provides a dynamic, user-friendly interface that allows trained observers to log standardized metadata, media characteristics, and contextual annotations in real time. Its modular form architecture organizes inputs across five categories: event selection, post metadata, content traits, user annotation, and optional media uploads — supported by conditional logic to reduce errors and ensure data integrity.
Trained researchers and volunteers manually input posts of interest using predefined classification values, minimizing ambiguity and free-text entry. Internal validation procedures, including inter-rater reliability checks and format validations, help ensure data quality, while outputs are securely stored and exported for integration into analysis workflows. Access is restricted to project team members, and the tool collects only
publicly available data in accordance with ethical research standards. A feedback loop and modular design allow for rapid iteration and adaptability to emerging research needs and narrative threats.
# **7.6. Public tipline**
The public tipline was created as a crowdsourced monitoring tool to help capture digital threats and misinformation incidents that often go undetected by automated systems or dominant online voices. To build an effective and scalable system, developers first conducted a series of interviews with analysts and experts to understand challenges such as inconsistent user input, information overload, and the need for rapid response. These insights informed the core design principles: accessibility, structure, scalability, and support for both structured and unstructured data. Using KoboToolbox and XLSForm technology, the form was designed to collect both free-text and categorical inputs, including incident details, source metadata, contextual tags, and uploaded media. Built-in logic and validation helped guide users and improve submission quality.
To streamline analysis, a triage framework was developed in parallel, allowing each submission to be scored based on factors like source credibility, public harm potential, relevance to known threat narratives, and quality of evidence. This system reduced review time and allowed analysts to prioritize cases efficiently. Data was securely stored in encrypted environments with strict access controls, and data minimization and retention policies were implemented to meet ethical standards. Following deployment, the tool was regularly reviewed and updated to address usability, emerging threats, and new incident types ensuring it remains a responsive and reliable system for digital threat monitoring.
# **8. Glossary**
**Astroturfing** occurs when individuals, companies, or organizations create the appearance of a grassroots movement supporting certain ideas or products, but the campaign is actually orchestrated and financed with targeted messaging.
**Avatars** are controlled representations designed to simulate distinct user profiles for the purpose of monitoring or auditing digital information flows. These representations are instrumental in studying how different user types experience online content and in assessing the health of the information ecosystem
**Bots** are software-driven user accounts that mimic real users online to amplify messages, boost engagement, or disseminate content, often as part of orchestrated campaigns.
**Deepfakes** are AI-generated or manipulated images, audio, or video that realistically depict people saying or doing things they never did, often used to mislead, deceive, or distort public perception.
**Disinformation** is deliberately false or misleading information created and spread with the intention to deceive and is often employed to manipulate public opinion, sow discord, or undermine democratic processes.
**Domestic manipulation** refers to efforts by actors within a country, such as political parties, influencers, or media outlets, to distort information, mislead the public, or amplify polarizing narratives for partisan or strategic gain. Unlike foreign interference, which originates abroad, domestic manipulation arises from internal actors and is often harder to distinguish from normal political campaigning.
**Election integrity** refers to the overall fairness, transparency, and trustworthiness of democratic processes. It encompasses both the technical security of election infrastructure and the reliability of the information environment in which citizens form opinions and make choices. Safeguarding election integrity involves protecting against manipulation, disinformation, and interference while ensuring equal access, accurate information, and public confidence in outcomes.
**Foreign influence** refers to overt efforts by foreign states or actors to shape another country's political processes, outcomes, or public opinion. This can include diplomatic pressure, lobbying, public messaging, or funding of diaspora institutions. It is conducted openly and does not necessarily break laws or norms.
**Foreign interference** refers to covert, deceptive, or clandestine activities by foreign states or their proxies intended to disrupt, manipulate, or undermine democratic institutions, elections, or public trust. This can include secret funding, cyber intrusions, intimidation of diaspora communities, or hidden disinformation campaigns.
**Generative AI** refers to artificial intelligence systems designed to create new content, such as text, images, audio, or video, that mimics human-produced material. These tools can be used for innovation and productivity, but also enable the rapid production of convincing false or misleading content, amplifying risks to the information ecosystem.
**Inauthentic platform activity** refers to coordinated or deceptive use of digital platforms, such as fake accounts, bots, or manipulated engagement metrics, to create a false impression of popularity, consensus, or legitimacy. This activity distorts online discourse by amplifying certain messages or suppressing others in ways that mislead users about genuine public opinion.
**Influencer** refers to an individual with the ability to shape opinions, behaviours, or consumption patterns of online audiences through their social media presence. Influencers can be authentic community voices or leveraged (often covertly) by campaigns and organizations to amplify messages and sway public opinion.
**Information ecosystem** refers to the interconnected system of people, platforms, institutions, and technologies involved in the creation, distribution, and consumption of information. It includes traditional media, social media, online communities, and interpersonal networks, and shapes how political claims and narratives spread, compete, and influence public opinion.
**Information Incident** refers to any coordinated or deliberate effort to manipulate public opinion, distort facts, or sow confusion during an electoral period. During the election, we monitored and triaged incidents into minor, moderate, or major depending on the level of disruption. They can involve suspicious, contentious, or malicious narratives, accounts, platform activities, and online behaviours.
**Malinformation** refers to information that is based on reality but is deliberately presented in a misleading or harmful way,often by taking facts out of context, amplifying them selectively, or releasing them at a strategic moment. Unlike misinformation (false but shared without intent) or disinformation (deliberately false), malinformation weaponizes genuine information to cause damage, undermine trust, or manipulate public opinion.
**Manipulated content** refers to genuine information, such as images, audio, or video, that has been altered, edited, or presented in misleading ways to distort its original meaning or context. This type of content exploits partial truths to deceive audiences and influence perceptions.
**Misinformation** refers to false or inaccurate information that is shared, generally without the intent to deceive. Unlike disinformation, which is deliberately created to mislead, misinformation often spreads unintentionally when people pass along incorrect claims, rumors, or misinterpretations.
**Nodes** refer to the individual entities in a network that can be connected to each other through edges; nodes in an information ecosystem are generally used to represent people or organisations.
**Online extremism** refers to the use of digital platforms and networks to promote, recruit for, or coordinate around extremist ideologies. It often involves spreading hateful or violent narratives, exploiting grievances, and building online communities that reinforce radical worldviews, sometimes leading to real-world harm.
**Platform regulation** refers to the policies, laws, and oversight mechanisms that govern how digital platforms operate, often in relation to content moderation, transparency, data practices, and accountability. In the electoral context, platform regulation seeks to reduce harms such as disinformation, inauthentic activity, and online extremism while balancing freedom of expression and democratic participation.
**Polarization** refers to the growing division of society into opposing ideological or political camps, where attitudes harden and common ground diminishes. In the information ecosystem, polarization is often intensified by targeted content, disinformation, and algorithmic amplification that reinforce echo chambers and deepen mistrust between groups.
**Transnational repression** refers to efforts by foreign states to intimidate, harass, or control diaspora and exile communities beyond their borders. This can include surveillance, online threats, disinformation, and coercion aimed at silencing dissent or discouraging political participation in the host country.
## **Contributors**
*Contributors to this report are:*
Aengus Bridgman,
Director
Jennie Phillips, Lead Researcher
Alexei Abrahams, Lead Researcher
Isabelle Corriveau,
Director, Public Outreach
and Strategy
Alexandra Stanghellini,
Junior Researcher
Ashley Vu,
Junior Researcher
Auxane Bussac, Junior Researcher
Ben Steel, Researcher
Chris Ross, Researcher
David Hobson, Researcher
Dinara Karimova, Researcher
Diya Jiang,
Researcher
Emma Frattasio,
Social Media Manager Esli Chan,
Researcher
Researcher
Fakhreddine Riahi,
Daniel Nikoula,
Researcher
Hana Allal,
Junior Researcher Junyan Zhu,
Researcher
Kathia Légaré, Researcher
Kiran Gill,
Junior Researcher
Liam Murphy, Junior Researcher
Luke Pindera, Junior Researcher
Mahnan Omar, Researcher
Mathieu Lavigne, Researcher
Mika Desblancs-Patel,
Researcher
Nader Hedfi, Researcher
Nishtha Gupta, Junior Researcher
Olivia Melanson Junior Researcher
Saewon Park, Researcher
Sara Parker, Researcher
Sarah Tran,
Social Media Manager
Spencer Izen, Junior Researcher
Victoria Forte, Junior Researcher
Zeynep Pehlivan, Researcher
Taylor Owen,
Principal Investigator
Skyler Ash Copy Editor During the election, the [Coalition on Informa](https://www.cdmrn.ca/coalition-for-information-ecosystem-resilience)[tion Ecosystem Integrity](https://www.cdmrn.ca/coalition-for-information-ecosystem-resilience) consisted of the following organizations:
**AI and Humanities Lab**
**Applied AI Institute**
**Apathy is Boring**
**Centre for the Study of Democratic Institutions**
**Chaire UNESCO-PREV**
**Democratic Engagement**
**Exchange**
**DFRLab**
**Digital Democracies Institute**
**Digital Public Square**
**Digital Society Lab**
**DisinfoWatch**
**Goodbot**
**Information Integrity Lab**
**MediaSmarts**
**Mila AI Institute**
**OpenMedia**
**Pol Comm Tech Lab**
**Reset Tech**
**Samara Centre for Democracy**
This report was made possible by the considerable work of a large number of collaborators, partners and researchers in Canada.
We thank Brian Morgan for his design work.
We also thank all Canadians who engaged with our work throughout the election and especially all those who submitted tips to our tipline.
## **Endnotes**
- **1 An avatar is defined as a virtual representation of users in an artificual environment, commonly used in the digital world (Sah et al., 2021).**
- **2 Since Meta disabled their CrowdTangle API and platform on August 14, 2023 researchers have not been able to systematically collect data on posting behaviour from public figures in a way consistent with the Observatory's cross-platform comparison approach. Facebook continues to be the most used platform by Canadians, as shown in Figure 1, and is very heavily used by Canadian politicians as well. This lack of visibility is deeply unfortunate and limits collective visibility into the dynamics and manipulations of the information ecosystem.**
- **3 This includes sports and entertainment news outlets. If we remove these, the engagement received by the top five news outlets jumps to 72%.**
- **4 Automated gender classification was applied using OpenAI's GPT-4.5 language model, based on candidate names. This approach necessarily reduces gender to a binary classification and may misclassify non-binary and gender-diverse candidates.**
- **5 PageRank is a link analysis algorithm and it assigns a numerical weighting to each element of a hyperlinked set of documents with the purpose of "measuring" its relative importance within the set.**
- **6 Based on a controlled 15-minute scroll per avatar across major platforms (Instagram, X, Facebook, TikTok, RedNote, and WeChat).**
- **7 Note, imbalance in viewing pro-Palestinian versus pro-Isreali narratives is likely due to our design limitations: Although we included a Jewish avatar, it was not as active as others, resulting in a more left-leaning observation overall.**
- **8 In line with the growing academic consensus and given the difficulties with attribution of intent, we use misinformation as an umbrella term to describe both intentional attempts to mislead and good faith production and sharing of false or misleading information.**
- **9 We have consistent coverage of all platforms except for Bluesky, which was recently added to our dataset and only includes candidates and journalists. Since our search relied on a keyword based approach, it represents a lower-end estimate of the fraud claims.**
- **10 The majority (76%) of the claims made by PPC candidates did not fall into the main topics identified. Many focus on claims of unfair treatment of their party leader, Bernier, by Elections Canada and the Leaders Debate Commission.**
- **11 2026 Alberta referendum: Premier Danielle Smith's government passed legislation lowering the threshold for citizen-initiated referendums, requiring signatures from 10% of eligible voters (≈177,000) instead of 20% to trigger a vote. While Smith opposes separation, her administration pledged to honor petitions meeting the new criteria, potentially allowing a sovereignty question on the 2026 ballot. Indigenous leaders from Treaties 6, 7, and 8 vowed legal challenges, arguing separation would violate constitutional rights.**
- **12 RedNote, also known as Xiaohongshu is a popular Chinese e-commerce and lifestyle social media platform. The platform integrates user-generated content with Plog, vlog and live-streaming features. The platform is particularly known for its popularity among female users (approximately** 70% **of its audience) and among oversea Chinese communities.**
- **13 Our investigation examined 11 different accounts, however we have observed more accounts sharing similar features.**
- **14 RedNote employs a content-based ranking system in pushing for content, putting weight on the quality of the content itself rather than overall follower counts of the account. As a result, this algorithm can allow content shared by accounts with few followings to reach a broad audience if it's deemed as quality content.**