**MEDIA ECOSYSTEM OBSERVATORY BRIEF**
# **Conspiratorial Claims and Institutional Distrust in Canada's Online Ecosystem**
February 2026
# **Contents**
| Executive summary
1 |
|------------------------|
| Résumé exécutif
1 |
| Context
4 |
| Questions
5 |
| Conclusion
13 |
| Methodology15 |
| Appendix
17 |
| Endnotes
19 |
| Bibliography
19 |
| Contributors
20 |
# **Conspiratorial Claims and Institutional Distrust in Canada's Online Ecosystem**
# **Executive summary**
Public critique of democratic institutions can often be a sign of a healthy democracy, ensuring greater institutional transparency and accountability. However, they can also be a destabilizing force against democratic structures and social cohesion. While Canada has maintained relatively high levels of institutional trust [over the past 15 years](https://web.archive.org/web/20260117095626/https://www.environicsinstitute.org/projects/project-details/trust-in-political-institutions-in-canada-(2025)), distrust is increasing among some segments of the electorate. Conspiracy claims can be conceived as a consequence of this growing distrust, but also cause more distrust. This is a vicious circle. These claims have had tangible consequences, shaping [public](https://doi.org/10.37016/mr-2020-028) [compliance with COVID-19 public health measures1](https://doi.org/10.37016/mr-2020-028) and [eroding](https://meo.ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election) [confidence in the integrity of democratic elections](https://meo.ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election).2 Today's information environment, which rewards attention and engagement, can amplify these claims and entrench them in the public debate.
This brief evaluates the spread and belief of anti-institutional conspiratorial claims in the Canadian information ecosystem. We define 'anti-institutional conspiratorial claims' as narratives that transform institutional skepticism into claims of covert elite coordination and harm, often detached from factual
# **Discours conspirationnistes et défiance institutionnelle dans l'écosystème canadien en ligne**
# **Résumé exécutif**
La critique publique des institutions démocratiques peut souvent être le signe d'une démocratie en bonne santé, en favorisant une plus grande transparence institutionnelle et une meilleure reddition de comptes. Cependant, elle peut aussi devenir une force déstabilisatrice pour les structures démocratiques et la cohésion sociale. Bien que le Canada ait maintenu des niveaux relativement élevés de confiance institutionnelle [au cours des](https://web.archive.org/web/20260117095626/https://www.environicsinstitute.org/projects/project-details/trust-in-political-institutions-in-canada-(2025)) [15 dernières années](https://web.archive.org/web/20260117095626/https://www.environicsinstitute.org/projects/project-details/trust-in-political-institutions-in-canada-(2025)), la méfiance augmente au sein de certains segments de l'électorat. Les discours conspirationnistes peuvent être perçus comme une conséquence de cette méfiance croissante, mais ils contribuent aussi à l'alimenter. Il s'agit d'un cercle vicieux. Ces discours ont eu des effets concrets, [influen](https://doi.org/10.37016/mr-2020-028)[çant l'adhésion du public aux mesures de santé publique liées à la](https://doi.org/10.37016/mr-2020-028) [COVID-19](https://doi.org/10.37016/mr-2020-028) et [érodant la confiance envers l'intégrité des élections](https://meo.ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election) [démocratiques](https://meo.ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election). L'environnement de l'information actuel, qui récompense l'attention et l'engagement, peut amplifier ces discours et les ancrer dans le débat public.
Ce court rapport évalue la diffusion et l'adhésion à des discours conspirationnistes anti-institutionnels dans l'écosystème canadien de l'information. Nous définissons les « discours conspiraevidence. To evaluate these claims, we draw on two sources of evidence:
- 1. A *nationally representative survey* measuring Canadians' awareness and belief in conspiratorial claims
- 2. A dataset of 14 million *posts from relevant accounts in Canada* on X, TikTok, Instagram, and Bluesky, used to analyze the content, origins, and spread of conspiratorial claims.
#### **Key Takeaways**
- E **Dominant conspiratorial claims challenge the legitimacy of democratic institutions:** High-engagement theories portray media, political elites, and public authorities as acting in coordinated ways to manipulate the public, elections, and health information, undermining trust in institutional processes rather than individual actors.
- E **Awareness is widespread, but belief remains limited:** Between 29% and 63% of Canadians report hearing about the conspiracy theories included in our survey, but only a small minority endorse them. Support is slightly higher for claims about gender indoctrination and media–elite collusion.
- E **Influencers drive both the supply and amplification of conspiratorial claims:** Influencers shared 87% of the conspiratorial claims we identified and attracted 89% of views and 87% of likes, highlighting their role in shaping attention to these claims.
- E **Platform dynamics shape exposure and belief:** X is the dominant platform for engaging with conspiracy theories, accounting for 70% of likes and hosting the largest volume of posts. Frequent users of X are also significantly more likely to be aware of and endorse conspiratorial claims
tionnistes anti-institutionnels » comme des récits qui transforment le scepticisme à l'égard des institutions en allégations de coordination secrète entre élites visant à nuire à la population, souvent détachées de toute preuve factuelle. Pour évaluer ces discours, nous nous appuyons sur deux sources de données :
- 1. un *sondage national représentatif* mesurant la connaissance et l'adhésion des Canadien·ne·s à des discours conspirationnistes;
- 2. un ensemble de données de 14 millions de publications provenant de *comptes pertinents au Canada* sur X, Tik-Tok, Instagram et Bluesky, utilisé pour analyser le contenu, les origines et la diffusion de ces discours.
#### **Principaux constats:**
- E **Les discours conspirationnistes dominants remettent en cause la légitimité des institutions démocratiques :** Les théories suscitant le plus d'engagement présentent les médias, les élites politiques et les autorités publiques comme agissant de manière coordonnée pour manipuler le public, les élections et l'information en santé, sapant ainsi la confiance envers les processus institutionnels plutôt qu'envers des acteurs individuels.
- E **L'exposition aux théories du complot est répandue, mais l'adhésion demeure limitée :** Entre 29 % et 63 % des Canadien·ne·s déclarent avoir entendu parler des théories du complot incluses dans notre sondage, mais seule une petite minorité y adhère. Le soutien est légèrement plus élevé pour les discours portant sur l'endoctrinement lié au genre et la collusion entre médias et élites.
- E **Les influenceur·euse·s alimentent à la fois l'offre et l'amplification des discours conspirationnistes :** Les influenceur·euse·s ont partagé 87 % des discours conspirationnistes que nous avons identifiés et ont généré 89 % des vues et 87 % des mentions « J'aime », ce qui met en évidence leur rôle dans la diffusion de ces discours.
- E **Les dynamiques propres aux plateformes façonnent l'exposition et l'adhésion :** X est la plateforme dominante en matière d'engagement envers les théories du complot, représentant 70 % des mentions « J'aime » et hébergeant le plus grand volume de publications. Les utilisateur·rice·s fréquent·e·s de X sont aussi nettement plus susceptibles d'être au courant des discours conspirationnistes et d'y adhérer que les utilisateur·rice·s d'autres plateformes ou les utilisateur·rice·s peu actif·ve·s sur les médias sociaux, qui sont les moins susceptibles d'en avoir connaissance et d'y croire.
- than users of other platforms or infrequent social media users, who are the least likely to be aware and believe in conspiratorial claims.
- E **A small group of users drives interconnected conspiratorial claims:** 100 online users account for 68% of conspiratorial posts, capturing 90% of views and 86% of likes. The same users generally promote multiple conspiratorial claims, particularly about the deep state, election fraud, and media-elite collusion, and gain attention primarily by posting around high-profile events.
#### **Why it Matters**
Anti-institutional conspiratorial claims can distort public debate, polarize citizens, and influence policy discussions. While some draw on genuine anxieties or partial truths, they often simplify complex issues into fear-driven narratives that normalize distrust in democratic institutions and media and shift attention away from evidence-based decision-making. Understanding how these claims gain traction online, and who engages with them, is essential for strengthening an information ecosystem that supports public accountability, high-quality information sources, and mutual (earned) trust between governments and citizens.
E **Un petit groupe d'utilisateur·rice·s est à l'origine d'un réseau de discours conspirationnistes interconnectés :** 100 utilisateur·rice·s en ligne sont responsables de 68 % des publications conspirationnistes, générant 90 % des vues et 86 % des mentions « J'aime ». Ces mêmes comptes promeuvent généralement plusieurs discours conspirationnistes, en particulier ceux liés à l'« État profond », à la fraude électorale et à la collusion entre médias et élites, et captent l'attention surtout lorsqu'ils font des publications autour d'événements très médiatisés.
#### **Pourquoi c'est important**
Les discours conspirationnistes anti-institutionnels peuvent déformer le débat public, polariser la population et influencer les discussions sur les politiques publiques. Bien que certains s'appuient sur des inquiétudes réelles ou des éléments partiellement véridiques, ils réduisent souvent des enjeux complexes à des récits alarmistes qui normalisent la défiance envers les institutions démocratiques et les médias, et détournent l'attention de la prise de décision fondée sur des données probantes. Comprendre comment ces discours prennent de l'ampleur en ligne — et qui y adhère — est essentiel pour renforcer un écosystème canadien de l'information qui soutient la reddition de comptes, l'accès à des sources d'information de qualité et une confiance mutuelle (méritée) entre les gouvernements et les citoyen·ne·s.
Issey Bailey/Unsplash
# **Context**
Canadians' trust in democratic institutions has remained comparatively high and stable [over the past 15 years](https://web.archive.org/web/20260117095626/https://www.environicsinstitute.org/projects/project-details/trust-in-political-institutions-in-canada-(2025)). Yet trust varies across groups, with some Canadians showing growing skepticism toward institutions such as the [media3](https://www.cem.ulaval.ca/wp-content/uploads/2025/06/dnrcanada2025e.pdf), [scientific commun](https://www.nature.com/articles/s41562-024-02090-5)[ity](https://www.nature.com/articles/s41562-024-02090-5)4, and [Elections Canada5](https://www.cdmrn.ca/publications/the-canadianinformation-ecosystem-during-the-2025-federal-election). While skepticism towards institutions is not inherently problematic, and can in fact encourage democratic accountability and transparency, its combination with conspiratorial claims of covert elite coordination, harm, and deception can undermine the legitimacy of democratic institutions. When institutional skepticism converges with conspiratorial claims, it can be weaponized – both domestically and [by foreign actors6](https://www.cdmrn.ca/publications/foreign-state-creators-response-to-charlie-kirk-assassination) — to shape policy preferences, amplify distrust and destabilize democratic governance.
Conspiracy theories serve several political, social, and psychological functions that enhance their appeal and make them difficult to challenge. Politically, conspiracy theories are fueled by skepticism inherent in anti-institutionalism, drawing on genuine concerns about transparency, accountability, and inequality that contain elements of truth and resonate with the public. It features an 'us versus them' divide between everyday citizens and political and economic elites, particularly resonant within rising economic and social inequalities in today's political climate.7 Socially, conspiratorial claims foster a sense of belonging by bringing people together around a shared awareness of a perceived "hidden truth" or collective injustice.8 Psychologically, conspiracy theories tend to thrive during times of crisis, as they help reduce anxiety and restore a sense of control amidst uncertainty by offering coherent explanations for complex events.9 This grounding in real concerns and lived experiences makes conspiracy theories feel plausible and personal to many, complicating efforts to challenge or address them.
The current online information environment provides a fertile ground for anti-institutionalist conspiracy theories to be created and shared. Increasing online segmentation, [the reduced](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada) [visibility of factual news on social media](https://meo.ca/work/old-news-new-reality-a-year-of-metas-news-ban-in-canada)10, and the [growing prom](https://meo.ca/work/power-shift-the-rise-of-political-influencers-in-canada)[inence of decentralized content creators and personality-driven](https://meo.ca/work/power-shift-the-rise-of-political-influencers-in-canada) [commentary](https://meo.ca/work/power-shift-the-rise-of-political-influencers-in-canada)11 reduce information gatekeeping and amplify emotionally engaging, identity-consistent content, creating conditions in which conspiratorial claims about public institutions can reach large audiences. These claims align closely with platform dynamics that prioritize engagement and controversy, creating economic incentives for their production and circulation. The advent of generative AI further lowers the cost and skill barriers to creating and disseminating persuasive conspiratorial claims at scale.
Against this backdrop, this brief defines 'anti-institutional conspiratorial claims' as narratives that transform institutional skepticism into claims of covert elite coordination and harm, often detached from factual evidence. We evaluate the prevalence of anti-institutional conspiratorial claims in the Canadian information ecosystem and map how these claims spread, their prevalence on social media platforms, identify the key channels and actors involved in their dissemination, and assess their uptake among the broader Canadian public. Our goal is to generate a better understanding of how these claims travel and shape public perceptions, helping inform efforts to curb their negative impacts on Canadian society and democracy.
#### **Limitations**
The internet is a complex environment in which language is often used figuratively, conveying different meanings to different audiences. While our main research focus in this brief is on claims that attribute covert and malicious intent to powerful or hidden actors, our methods are not perfectly capable of capturing the nuances of language in specific contexts, a challenge that humans also struggle with.
Conspiratorial-coded language can be used in contexts that would not fit our definition. For example, Dominic Cummings, an advisor to former Prime Minister Boris Johnson, casually referred to the "deep state" in the context of the UK bureaucracy preventing the transition to Microsoft Teams or Google Docs [during a podcast](https://youtu.be/3i7ym_Qh7BA?si=WIt2im0a8ykCPYdu&t=196) discussing the inner workings of government. This language clearly references the deep state narrative, but in a sanitized manner, devoid of the malintent presupposed by our definition of anti-institutional conspiratorial claims. Therefore, we acknowledge the limitations of our methods in distinguishing these contexts, and transparently report our definitions, methods, and data to help the reader understand how we reached our conclusions.
# **Questions**
In this brief, we use claim extraction and stance analysis to evaluate the prevalence of a range of conspiratorial claims in the Canadian information ecosystem, alongside a nationally representative survey to measure their salience amongst the Canadian public. We specifically look at the following questions:
### **1. What are the main themes and arguments associated with these conspiratorial claims?**
We identify the main themes and arguments by applying topic modelling and named entity recognition to social media posts supporting each conspiratorial claim across four platforms and by qualitatively analyzing the 20 posts that received the most likes for each conspiracy. This analysis provides an overview of what conspiratorial claims are about before evaluating their prevalence and origin.
#### **2. How prevalent are these conspiracy theories in the Canadian information ecosystem?**
We use a nationally representative survey to evaluate Canadians' awareness and belief in each conspiratorial claim. We also examine the total number of views and likes that posts from Canadian users promoting each conspiratorial claim have received and track how engagement with these conspiracies has changed over time.
## **3. Who shows the highest exposure to and belief in these conspiratorial claims?**
Based on a nationally representative survey, we examine how respondents' awareness of and belief in conspiratorial claims vary by age, gender, and use of different social media platforms.
### **4. Who is spreading these conspiratorial claims, and on which platforms are they spreading?**
We assess the roles of politicians, influencers, media organizations, civil society groups, and government organizations in the spread of conspiratorial claims in the Canadian information ecosystem. We also investigate whether the same actors promote multiple conspiracies and the degree to which the spread is concentrated among a small number of users. Finally, we analyze how engagement with conspiratorial claims varies across social media platforms.
To answer those questions, we combine a nationally representative survey of Canadians with social media data from four platforms (TikTok, X, Instagram, and Bluesky). The survey was administered to 1,459 Canadians between October 30th and November 7th, 2025. We sampled and weighted our data by age, gender and province, according to population estimates provided by Statistics Canada. The social media data draw on X, TikTok, Instagram and Bluesky posts (14,454,308 posts), and engagement statistics for those posts (views and likes), from accounts in our comprehensive seedlist of news organizations, politicians, influencers, and civil society and government organizations (4,501 entities) relevant to the Canadian information environment between January 1st, 2023 and September 1st 2025. Of those posts, 69,201 were identified as supporting at least one of the eight conspiracy theories listed below. More details on data collection and how supporting posts were identified are provided in the Methods section.
## **Question 1:**
# **What are the main themes and arguments associated with these conspiratorial claims?**
Anti-institutional and conspiratorial claims do not exist in isolation; they are part of broader public narratives, political grievances, and cultural touchpoints. To evaluate the prominence and characteristics of these claims we identify eight that are particularly prominent and relevant to the Canadian context and have the potential to undermine trust in key areas of public life, including public health, education, media, elections, technologies, and climate policy. These conspiratorial claims were selected based on a qualitative review of existing literature and trends in the online information ecosystem. Although they do not represent the full spectrum of conspiratorial content circulating in Canada, they encompass a range of issues and interests relevant to contemporary Canadian society.
- 1. **Health threats**: Public health threats (e.g. COVID-19) are exaggerated to expand government control.
- 2. **Gender indoctrination**: Schools are indoctrinating kids with radical gender ideology.
- 3. **Media-elite collusion**: Major Canadian media outlets conspire with political elites to manipulate public opinion.
- 4. **Election fraud:** Vote counts are faked during elections in Canada.
- 5. **Deep state:** A secretive group of elites ('Deep State') holds most of the real power in Canada.
- 6. **Digital ID**: Digital IDs are used by the government to secretly control Canadians.
- 7. **Intentional wildfires**: Wildfires in Canada are intentionally started to advance the eco-agenda.
- 8. **Climate hoax:** Canada fakes environmental data to expand government control over Canadians.
To better understand the themes and arguments they entail, we qualitatively analyzed the 20 most-liked posts associated with each identified claim. We also applied topic modeling, named entity recognition and noun phrase detection (see Methodology section for details) to all 69,201 posts supporting any of the eight theories to identify key discussion topics for each. The following summarizes the findings from these analyses. For plots of the top topics, named entities, and noun phrases for each conspiracy, see Figures A1-A3 in the Appendix.
#### **Health threats:**
#### *Public health threats (e.g. COVID-19) are exaggerated to expand government control*
Claims that the COVID-19 pandemic was planned in advance and weaponized to enable authoritarian control. Many posts portray vaccines as unsafe, not properly tested, or intentionally harmful, with several framing both the virus and vaccination efforts as coordinated biowarfare and genocide by elites. Many of these posts implicate Bill Gates, claiming he controls the World Health Organization (WHO) and pushes a global vaccine agenda to benefit financially from vaccination efforts. Others promote distrust in public health institutions, mention government fear-mongering over the risks of COVID-19, or claim that immigrants are massively leaving Canada because of the construction of "[COVID camps](https://web.archive.org/web/20251230192824/https://www.cbc.ca/news/politics/covid-19-internment-camps-disinformation-1.5769592)," where Canadians would be forcibly sent to quarantine.
# **Gender indoctrination:**
## *Schools are indoctrinating kids with radical gender ideology*
Claims that schools are promoting "radical gender ideology" without parental consent or knowledge through curricula like the Sexual Orientation and Gender Identity (SOGI) teaching resources, drag queen story times, and transgender medical interventions for minors. Key concerns include alleged restrictions on parental rights, government overreach, and protecting students who are framed as indoctrinated, fearful, confused, or losing their innocence. One post also criticizes children's TV shows and movies, particularly at Disney, for having a "woke" agenda which "many parents see as an attempt to indoctrinate their children". Among these posts, the phrases "women" and "girls" are mentioned 33% more often than "men" and "boys", indicating a greater rhetorical concern over the education, protection, and social environment of women and young girls.
# **Media-elite collusion:**
#### *Major Canadian media outlets conspire with political elites to manipulate public opinion*
These posts claim that mainstream media are controlled, biased, or captured by powerful actors, primarily because of the funding they receive from government or industry. They largely contend that news organizations serve the government and corporate agendas, acting as mouthpieces for the Prime Minister's Office, the World Economic Forum (WEF), or other elites and institutions. Some portray mainstream Canadian outlets, including the CBC and CTV, as state-controlled media. Claims of mainstream media as "fake news" are prevalent, including arguments to defund the CBC.
# **Election fraud:**
#### *Vote counts are faked during elections in Canada*
Many posts advance claims of election fraud related to the 2020 and 2024 US presidential elections. Of the posts concerning Canadian elections, many portray the longest ballot in Pierre Poilievre's riding during the 2025 election as an illegal scam, or claim that China helped Justin Trudeau win the 2021 election. Other posts focused on specific dimensions of the voting process in Canada, including [the need for voters to bring their own pen1](https://meo.ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election)2 rather than using the provided pencil when voting to ensure that their vote cannot be altered. Rampant corruption among political elites is a common theme within these posts.
**Figure 1: Awareness, belief, and engagement with eight conspiracy theories.**
#### **Deep state:**
## *A secretive group of elites ('Deep State') holds most of the real power in Canada*
A large number of posts focus on the period when Justin Trudeau resigned, and Mark Carney became prime minister without having been elected by the general public. These posts generally frame Mark Carney's leadership as illegitimately installed by the liberal establishment or global elites such as the WEF, or denounce his ties with the Chinese government. More generally, posts present Canadian democracy, including both Justin Trudeau and Pierre Poilievre, as effectively held "hostage" by elite interests, citing policies favouring their wealthy friends, cover-ups of corruption scandals, unjustified prorogation of parliament, and the use of emergency powers as evidence. The WEF features prominently in these posts, with accusations that it orchestrates global politics to serve wealthy elites rather than ordinary Canadians.
#### **Digital ID:** *Digital IDs are used by the government to secretly control Canadians*
The posts argue that government‑backed digital identity systems, central bank digital currencies, and the shift away from cash are part of a broader plan to control the population. They argue that digitization will enable governments to monitor financial transactions, restrict purchases, travel, and access to healthcare, freeze accounts, and punish people for exceeding their carbon limits or for dissent. These policies are compared to the social credit system in China, linked to global elites and institutions (e.g., the WEF), and presented as steps toward a surveillance state, totalitarianism, slavery, or a "prison planet". In this context, the posts suggest that citizens are "waking up" and frame resistance as necessary to preserve personal freedom, privacy, and democracy.
#### **Intentional wildfires:** *Wildfires in Canada are intentionally started to advance the eco-agenda*
The posts claim that wildfires in Canada are deliberately set by "green terrorist" arsonists, some paid by the government, in order to support climate change narratives and push policies seen as exploitative, including the carbon tax. For example, multiple top posts suggest that the Liberal government asked firefighters to stand down while wildfires spread in Jasper so that they could blame the fires on climate change. Posts also criticize the government's forest management and the funding of CBC's "climate propaganda."
#### **Climate hoax:**
#### *Canada fakes environmental data to expand government control over Canadians*
These posts largely claim that human-made climate change does not exist or that its severity is overstated, using words like "hoax," "scam," "hysteria," or "fearmongering." The posts suggest that climate change is used as a pretext to suppress freedom, destroy the economy, steal from Canadians, and ruin their way of life. Climate policies and narratives are framed as forms of
government control over what people can eat, what cars they drive, and how much energy they consume. Blame is generally attributed to Justin Trudeau, former Environment Minister Steven Guilbeault, the WEF, and "communists."
#### **Question 2:**
# **How prevalent are these conspiratorial claims in the Canadian information ecosystem?**
Although online conspiracy theories often target specific communities or concerns—for instance, Canadians in Northeast BC may encounter theories linking wildfires to eco-terrorism or environmental sabotage—it is [less clear how these claims circulate](https://www.tiktok.com/@etymologynerd/video/7561869312261131534?lang=en) [and resonate at the national level](https://www.tiktok.com/@etymologynerd/video/7561869312261131534?lang=en). Conspiracies can appear highly specific, yet their reach may extend far beyond the communities where they originate. This section evaluates the prevalence of such conspiratorial claims across the Canadian information ecosystem by examining overall levels of awareness, belief, and engagement among Canadians. We also analyze how levels of engagement with each conspiratorial claim change over time, showing how certain conspiracies attract bursts of engagement around particular posts or events, while others sustain a more uniform level of interest.
Survey results (left) report self-reported awareness of and belief in each conspiracy. Social media data (right) show views (X and TikTok) and likes (X, TikTok, Instagram, Bluesky) for posts related to each conspiracy. An example reading, *"Among surveyed Canadians, 47% reported awareness of the 'Media-elite collusion' conspiracy and 16% also reported believing it. On social media, posts from influential Canadian political accounts supporting the claim received 2.57 billion views and 4.9 million likes."*
The survey data in **Figure 1** (left side) show Canadians' awareness and beliefs for each of the eight conspiracies we surveyed. We find moderate to high awareness, ranging from 29% (climate hoax) to 63% (health threats) of Canadians being familiar with the provided conspiracy theories.13 When survey respondents who were aware of the provided conspiracy theories were asked if they also believed in them, a smaller but still notable subset of Canadians responded to some level of belief (Figure 1 categorizes 'probably' and 'definitely' true responses as 'believe.'). Agreement is highest for the gender indoctrination narrative, endorsed by 21% of Canadians, or 40% of those aware of the conspiracy. Media-elite collusion follows, with 16% of Canadians expressing belief. The next most believed conspiracy was media-elite collusion, where 16% of Canadians report belief. The remaining theories showed agreement levels ranging from 8% to 13%. Although awareness of individual conspiracies
varies widely (29–63%), belief remains much more concentrated (8% and 21%), highlighting that widespread exposure does not translate directly into endorsement.
Turning to social media data (right side of Figure 1), the mediaelite collusion conspiracy clearly dominates in reach, amassing over 2.57 billion views, more than twice as many as the next most viewed conspiracies, health threats and election fraud, each with roughly a billion views. All other conspiracies lagged behind, receiving fewer than 700 million views, with digital ID and intentional wildfire conspiracies attracting the least attention at under 100 million views. Although smaller than the top-viewed conspiracies, these view counts still represent massive reach, showing that even more marginal theories can attract substantial attention online. Like counts do not seem to directly correlate with views. Instead, we see that conspiracies about health threats (6.66 million likes) and the deep state (5.99 million likes) attract the most engagement. Again, all tested conspiracies garnered large and generally comparable like counts, except for the digital ID control and intentional wildfire conspiracies, which received comparatively lower engagement.
Next, we evaluate how engagement received by conspiratorial posts on social media in Canada evolves over time. In **Figure 2**, we show the evolution of the weekly sum of likes over time per conspiratorial claim from January 1st, 2023, to September 1st, 2025. We color the graph by the total number of likes a conspiracy received across the entire timeframe, with lighter colors indicating higher overall engagement. We observe wide differences in how engagement with each conspiracy unfolds over time, with peaks generally driven by a few posts from a small number of users.
Some peaks are associated with large-scale external events. For example, the "Intentional wildfires" posts tend to peak around summer when wildfires feature heavily in news reporting, [with](https://web.archive.org/web/20260116164642/https://natural-resources.canada.ca/stories/simply-science/canada-s-record-breaking-wildfires-2023-fiery-wake-call) [its largest peak happening during Canada's record-breaking wild](https://web.archive.org/web/20260116164642/https://natural-resources.canada.ca/stories/simply-science/canada-s-record-breaking-wildfires-2023-fiery-wake-call)[fires in the Summer of 2023.](https://web.archive.org/web/20260116164642/https://natural-resources.canada.ca/stories/simply-science/canada-s-record-breaking-wildfires-2023-fiery-wake-call) "Election fraud" posts peaked in February 2023 after [reports by CSIS revealed Chinese foreign](https://web.archive.org/web/20251005163235/https://commissioningerenceetrangere.ca/fileadmin/commission_ingerence_etrangere/Documents/Preuves_et_Presentations/Preuves/COM0000060.pdf) [interference during the 2021 election](https://web.archive.org/web/20251005163235/https://commissioningerenceetrangere.ca/fileadmin/commission_ingerence_etrangere/Documents/Preuves_et_Presentations/Preuves/COM0000060.pdf), but also periodically during the US federal elections in 2024 and Canadian federal elections in early 2025. Finally, "Gender indoctrination" peaked in September of 2023 from a viral TikTok video interviewing a protestor at the [1 Million March For Children](https://web.archive.org/web/20251204192147/https://theconversation.com/how-the-parental-rights-movement-gave-rise-to-the-1-million-march-4-children-213842s://theconversation.com/how-the-parental-rights-movement-gave-rise-to-the-1-million-march-4-children-213842) protest against gender education.
An example reading, *"In the week of July 8th 2024, likes received by posts pushing the 'Climate hoax' conspiracy reached ~81,000 likes, the maximum amount of likes it received across all weeks."*
**Figure 2: Weekly engagement (likes) with conspiratorial claims.**
Other conspiracies reach peaks by capturing viral moments. For example, the "Digital ID" conspiracy peaked in February 2023 from two viral TikTok videos from a single user [falsely](https://web.archive.org/web/20250917212519/https://factcheck.afp.com/doc.afp.com.33AY3LX) [arguing that a healthcare funding plan was tied to Canada's digital](https://web.archive.org/web/20250917212519/https://factcheck.afp.com/doc.afp.com.33AY3LX) [ID program](https://web.archive.org/web/20250917212519/https://factcheck.afp.com/doc.afp.com.33AY3LX). "Media-election collusion" posts tend to receive constant engagement but went viral on TikTok and Twitter after a few influencers shared [a video of Poilievre](https://www.youtube.com/watch?v=8UKiL7remis) accusing a Canada Press reporter of being a "tax-funded mouthpiece to the PMO" and suggesting they were funded so Justin Trudeau could "buy support from the media". Finally, "Deep state" posts received constant attention but peaked in January 2025 when an alternative news outlet characterized the leadership transition from Justin Trudeau to Mark Carney as a move from "a World Economic Forum puppet to a World Economic Forum script writer."
Finally, we find that others regularly experience large peaks in attention. For example, "Climate hoax" and "Health threats" posts received sustained engagement throughout our period and aren't explicitly linked to a single large-scale event.
#### **Question 3:**
# **Who shows the highest exposure to and belief in these conspiratorial claims?**
To identify the groups most vulnerable to conspiracy theories in the Canadian context, we use survey data to examine how awareness and belief in conspiracy theories vary by respondents' gender, age, and social media use. For simplicity, we average exposure and beliefs across the eight conspiracies.
We use a self-reported weekly user threshold to classify survey respondents as platform users, which are non-exclusive categories. This threshold captures respondents who engage with the platform regularly while ensuring that a meaningful sample size is available for each platform usership category.
**Figure 3** summarizes group differences in awareness and belief in conspiracies, using a bullseye plot to compare each group to the average Canadian (centre of the plot). Of the 8 conspiracies we surveyed Canadians about, the average respondent was aware of 3.49 conspiracies and believed in 0.91. When looking at gender, we see that men are generally more aware of and believe in more conspiracies than women. Younger Canadians, aged 18-34, are aware of more conspiracies than their older counterparts, but believe fewer of them. Middle-aged Canadians, aged 35-54, are close to the Canadian average in awareness but believe more conspiracies than other age groups. And older Canadians are generally much less likely to have heard about the conspiracies, and have comparatively low levels of belief (similar to those of the 18–34).
Each point shows how a group differs from the Canadian average (center of the plot) in the number of conspiracies they are aware of (x-axis) and believe (y-axis). Appendix Table A1 presents the raw values. Social media users are defined as people who report using a platform at least weekly, and are not exclusive categories. An example reading, *"Canadians who are users of the social media platform X are both more aware of and believe in more conspiracies than the average Canadian."*
Social media habits are a major driver of both awareness and belief in conspiratorial claims. Respondents who report low social media usage (do not use any of the platforms on a weekly basis) are the least likely to be aware of or believe in conspiracy theories. In the opposite corner of the plot are X users, who show the highest levels of awareness and belief in our measured conspiracies. All other social media user groups are above the Canadian average for awareness, and all but Bluesky users are above the Canadian average for belief. These findings suggest that frequent social media use is associated with greater exposure to conspiratorial claims, but that the effects vary across
**Figure 4: Percentage of content and engagement produced/received by the top N supporting posters of conspiracies, compared to posters discussing related themes.**
platforms due to their distinct dynamics and affordances.
One plausible explanation is that users may gravitate towards X to find like-minded networks and users engaged in conspiratorial discourse, viewing the platform environment as more open to anti-institutionalism or the critique of political elites. Further research could also examine whether X's algorithms may amplify content that pushes conspiratorial claims, thereby increasing users' exposure to those theories.
#### **Question 4:**
# **Who is spreading these conspiratorial claims, and on which platforms are they spreading?**
While Sections 2 and 3 examined awareness, belief, and exposure among Canadians, these patterns ultimately depend on the structure of the information environment. In this section, we therefore turn from audiences to producers by examining who creates and disseminates these claims and where they circulate online.
**Table 1** shows the percentages of posts, views, and likes for conspiratorial claims, categorized by user type (influencers, news organizations, politicians, civil society organizations, government organizations) and platform (X, TikTok, Instagram, Bluesky). Starting with user type, we find that influencers produce the majority of conspiracy-related posts (87%) and receive the lion's share of engagement, accounting for 89% of views and 87% of likes. News outlets and politicians represent a smaller but non-negligible share of posts, views and likes, highlighting the role they can play in spreading and legitimizing conspiratorial claims. Looking more closely at the conspiratorial activity of news outlets, we find that alternative news media accounts are by far the most influential. They account for roughly 78% of posts and 92% of likes, with two organizations alone capturing over 80% the engagement received by news outlets.
An example reading, *"87.4% of the 73.6 thousand conspiratorial posts identified were posted by influencers."*
The platform distribution indicates that X is the primary site of engagement with conspiratorial posts, accounting for 70% of total likes across the four platforms. TikTok comes in second with 20% of total likes, while Instagram receives 9.1 % and Bluesky 0.6 %. X's dominance can partly be attributed to its large volume of posts, with over 92% of conspiratorial posts from our seedlist being published on the platform. However, looking more closely at the top 20 posts per conspiracy, we find that TikTok makes up 48% of likes while X only makes up 41% of the likes. This suggests that while the volume of conspiratorial content circulating on X is substantially larger than on other platforms, an individual TikTok video can have as much impact as an individual tweet.
**Table 1: Distribution of posting activity, likes, and views by actor type and platform (in %), with overall totals by metric.**
| User Type | | | |
|--------------------|-------|-------|-------|
| | Posts | Views | Likes |
| Influencer | 87.4 | 88.6 | 86.5 |
| News outlet | 4.7 | 2.5 | 8.6 |
| Politician | 6.4 | 8.2 | 4.3 |
| Civil society org. | 1.5 | 0.7 | 0.6 |
| Government org. | 0.0 | 0.0 | 0.0 |
| Platform | | | | |
|------------|-------|-------|-------|--|
| | Posts | Views | Likes | |
| X | 92.5 | 98.7 | 70.3 | |
| TikTok | 3.1 | 1.3 | 19.9 | |
| Instagram | 3.8 | N/A | 9.1 | |
| Bluesky | 0.6 | N/A | 0.6 | |
| | | | | |
| Raw totals | 73 6K | 5 8R | 25 QM | |
Research on misinformation diffusion has shown that a very small number of accounts can be responsible for a disproportionate share of false and misleading content. Building on this finding, we evaluate the extent to which a small subset of social media users generates most of the conspiratorial content and attracts the majority of engagement on Canadian social media.
**Figure 4** shows the distribution of posts, views, and likes across top spreaders of conspiratorial claims and thematically related content. The top 100 spreaders are responsible for 68% of conspiratorial posts and capture 90% of views and 86% of likes on conspiratorial content. This demonstrates that conspiratorial claims are driven by a small, highly active and visible subset of social media users who exert a disproportionate influence on the information environment. Comparing these distributions to thematically related content (i.e., posts discussing the same topics without necessarily advancing conspiratorial claims), distributions remain highly concentrated, though less so than for conspiratorial content, with the top 100 posters accounting for 50% of posts.
The "General" type is defined as posts supporting, refuting, or discussing themes related to each conspiracy. An example reading, *"The top 100 posters of conspiratorial claims account for 68% of conspiratorial content, whereas the top 100 posters of thematically related content account for 50% of thematically related content."*
We examine the extent to which the same users tend to advance different conspiratorial claims. Of all users who have posted at least one post advancing one of the eight conspiratorial claims, 59% promoted at least two of them. We next calculate correlations between the number of posts each user makes about different conspiratorial claims. This tells us whether users who are highly active in promoting one conspiratorial claim are also active in promoting others. We find a relatively strong correlation between conspiratorial claims (the median correlation is 0.4). In particular, the 'Deep state', 'Election fraud,' and 'Media-elite collusion' conspiratorial claims all exhibit correlations greater than 0.7, showing that users who post heavily about one of these claims are also very likely to post heavily about the others. The topic modelling results from Question 1 help further contextualize these correlations. Among the topics most commonly shared by the conspiracies are claims of widespread political corruption, governments striving for authoritarian control, and elites seeking to coerce the public through fear. These further illuminate the fundamental and deeply rooted theme of distrust in institutions that permeates all the conspiracies.
Mark Konig/Unsplash
# **Conclusion**
Endorsement of anti-institutional conspiratorial claims remains limited in Canada, but they occupy a visible position in the information ecosystem and structure how institutional distrust is articulated and circulated online. The highest-engagement claims we observe portray media organizations, political elites, and public authorities as coordinated actors manipulating elections, public health information, and public opinion, challenging the legitimacy of democratic institutions and potentially shaping how Canadians perceive and engage with democratic processes.
Canadians report relatively high awareness of conspiratorial claims, despite the highly concentrated supply of conspiratorial content. A very small number of very active users account for most conspiratorial posts and capture most of the engagement, with the same users generally promoting multiple conspiratorial claims. Among the four platforms studied, X accounts form a disproportionately large share of conspiratorial content, which likely reflects both platform-level amplification and the characteristics of its users. These structural dynamics, which allow conspiratorial content to circulate widely, pose a challenge but can also be leveraged to amplify credible information and corrective messaging.
As conspiratorial claims build on existing public skepticism, addressing these dynamics also requires greater effort to understand the underlying public grievances and concerns that motivate them. This requires ensuring that citizens have access to credible information, avenues to engage in democratic processes, and opportunities to hold institutions accountable. Political officials and government institutions should also strengthen transparent and inclusive functions in their public engagement.
## **Recommendations**
The findings of this brief reveal that anti-institutional conspiratorial claims, while believed by a minority of Canadians, occupy a disproportionately visible position in Canada's information ecosystem. This visibility stems from structural features of the digital environment: a small number of highly active users generate the majority of conspiratorial content, platform algorithms amplify engagement-driven narratives, and influencers command outsized reach compared to traditional institutional actors. These patterns suggest that effective policy interventions must address both the supply-side dynamics that enable conspiratorial content to circulate widely and the demand-side factors, including legitimate institutional skepticism and information gaps, that make such claims resonate with segments of the public.
The policy recommendations outlined below operate across multiple levels of the information ecosystem. They target the structural conditions that allow conspiratorial claims to gain visibility, strengthen the credibility and accessibility of institutional information, and recognize that in an influencer-driven environment, reaching audiences increasingly requires working through the same creators and platforms where conspiratorial content thrives. The goal is to ensure that public discourse is grounded in evidence, that institutions operate transparently, and that Canadians have the tools to distinguish legitimate skepticism from unfounded conspiracy theories.
# **Recommendation #1: Targeted Platform Accountability**
- E Require platforms to provide transparency reports on amplification algorithms, including disclosure of how recommendation algorithms prioritize content. Reports should detail the signals and weights used in content ranking, the frequency of algorithmic updates, and any special treatment afforded to particular content types or accounts. Importantly, reports should be standardized across platforms and publicly accessible to enable independent research. The EU's Digital Services Act provides a model for such transparency requirements, mandating that "very large online platforms" submit annual risk assessments and provide researchers with data access under [Article 40.](https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A32022R2065)
- E Require impact assessments before major algorithmic changes that could affect information quality. Before implementing significant algorithmic or policy changes that affect content visibility or amplification, platforms should conduct and publish impact assessments documenting the processes used to identify, mitigate, and monitor risks, such as the spread of misleading or low-credibility content.
- E Strengthen user control over algorithms by enabling users to edit their personalization preferences. Users should have meaningful control over the content they see, including the ability to adjust algorithmic parameters, opt out of recommendation features, or switch between feed types. Similar user empowerment provisions are being implemented under the EU's Digital Services Act, which requires platforms to offer users at least one recommendation option [not based on profiling](https://digital-strategy.ec.europa.eu/en/policies/digital-services-act).
- E Improve researcher access to platform data. Platforms should provide researchers with secure, privacy-preserving access to data on content distribution, user engagement patterns, coordinated and inauthentic activity, and algorithmic decision-making.
# **Recommendation #2: Pre-Bunking and Proactive Disclosure**
- E Launch proactive public education campaigns during predictable conspiracy "peak seasons" (wildfire season, election periods, public health emergencies). These campaigns should employ pre-bunking strategies that familiarize audiences with common conspiratorial techniques, and should be timed to reach audiences before conspiratorial claims gain momentum. Pre-bunking approaches have shown promise internationally, with the UK's [Rapid](https://www.gov.uk/government/news/government-cracks-down-on-spread-of-false-coronavirus-information-online) [Response Unit](https://www.gov.uk/government/news/government-cracks-down-on-spread-of-false-coronavirus-information-online) and the European Commission's [pre-bunking](https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/misinformation-and-disinformation-both-prebunking-and-debunking-work-fighting-it-2024-10-25_en) [campaigns](https://joint-research-centre.ec.europa.eu/jrc-news-and-updates/misinformation-and-disinformation-both-prebunking-and-debunking-work-fighting-it-2024-10-25_en).
- E Strengthen proactive transparency mechanisms to address legitimate concerns that common conspiracies exploit and to enable greater institutional trust (i.e., government funding models, decision-making processes, budget transparency, institutional relationships, etc.). This disclosure should provide clear, accessible explanations that preemptively address the information gaps that conspiracies exploit.
## **Recommendation #3: Create Opportunities for Creator– Institution Engagement**
E Design frameworks to include independent digital creators in public information and pre-bunking campaigns. Governments, independent agencies, and civil society organizations should consider engaging with digital creators as part of public information, media literacy, and pre-bunking initiatives. Programs should be guided by clear, publicly articulated inclusion criteria and can take pilot or iterative forms, such as voluntary briefings or educational forums on high-salience civic issues.
# **Methodology**
#### **Survey Data**
To understand the general public's awareness and attitudes toward conspiracies, we fielded an original survey to a nationally representative sample of 1,459 Canadians, between October 30 and November 7, 2025. The margin of error for a comparable probability-based random sample of the same size is +/-2.6%, 19 times out of 20.
Survey questions used in analysis:
#### *Conspiracy Awareness and Belief*
#### **Awareness**
The following contains a list of events and claims made on social media. We would like to know which items you might have heard of and which ones you have not heard of.
(Yes, I have heard of this or something similar, No, I have not heard of this)
- E Public health threats (e.g. COVID-19) are exaggerated to expand government control.
- E Schools are indoctrinating kids with radical gender ideology.
- E Major Canadian media outlets conspire with political elites to manipulate public opinion.
- E Vote counts are faked during elections in Canada.
- E A secretive group of elites ('Deep State') holds most of the real power in Canada.Digital IDs are used by the government to secretly control Canadians.
- E Wildfires in Canada are intentionally started to advance the eco-agenda.
- E Canada fakes environmental data to expand government control over Canadians.
#### **Belief**
Please rate the extent to which you believe each statement below is true or false.
(Definitely false, Probably false, Not sure or undecided, Probably true, Definitely true)
(same claims as above shown to those who were aware of them)
#### **Social Media Use**
How often do you use each of the social media apps and websites listed below (for any reason)?
(Never, once or twice a month, several times a month, once or twice a week, several times a week, once or twice a day, several times a day)
E Facebook E X (formerly Twitter)
E YouTube E Instagram E TikTok E Bluesky
#### **Identifying conspiratorial posts from social media data**
Our starting dataset consisted of all social media posts on Instagram, X (formerly Twitter), Bluesky, Instagram and Tiktok from a comprehensive seed list of Canadian politicians, news outlets, influencers, civil society organizations, government organizations, and politicians relevant to the Canadian information ecosystem14 (see [Pehlivan et al., 2025\)](https://arxiv.org/abs/2506.10942) between January 1st, 2023 and September 1st 2025. A seed entity must be active on at least one platform and is defined as a person, group, organization, or media product of substantive interest to the ecosystem. In total, the seed list comprises 10,991 unique entities. Data was collected using various methods, including official APIs for platforms like Bluesky, and custom web scraping for platforms like X, Instagram and TikTok. Since TikTok is primarily a videofirst platform, we transcribed all posts. In total, we collected 14,454,308 posts. While creating a fully comprehensive dataset is not possible due to the dynamic and constantly evolving nature of social media, this dataset provides a representative snapshot of the Canadian information ecosystem.
#### *Claim extraction*
We used a small LLM (Qwen/Qwen3-8B) and used a few-shot prompt to extract every claim made in an X post or TikTok transcript. If no statements or claims are found, the model is instructed to return an empty list of claims.
What do we mean by claims? We view the process of extracting claims as extracting all atomic units within a social media post or transcript that express a single stated opinion or assertion about the world. For example, on September 18, former US President Barack Obama posted on X:
*"After years of complaining about cancel culture, the current administration has taken it to a new and dangerous level by routinely threatening regulatory action against media companies unless they muzzle or fire reporters and commentators it doesn't like."*
This post can be broken down into the following claims:
- 1. The current administration has complained about cancel culture for years.
- 2. The current administration has taken cancel culture to a new and dangerous level.
- 3. The current administration routinely threatens regulatory action against media companies.
- 4. The current administration threatens regulatory action against media companies unless they muzzle or fire reporters and commentators it doesn't like.
In total, we extracted 28,633,939 unique claims.
#### *Embeddings*
We used [Qwen/Qwen3-Embedding-0.6B](https://huggingface.co/Qwen/Qwen3-Embedding-8B) to generate embeddings for each claim of dimension 1,024. The motivation behind generating the embeddings of the claims instead of the entire text is to help with claim retrieval by avoiding drowning out claims that might occur in multiple posts or transcripts with surrounding text that might contain different or unrelated content. This process is especially important for longer texts like TikTok transcripts where multiple claims are made in a single post and transcripts might contain many different ideas.
#### *Entailment*
We frame the process of finding posts pushing a conspiracy as a *natural language inference* task where, given a post (e.g. an Instagram caption, a Tiktok description & transcription, etc.), and a claim, we try to find whether the claim follows from the textual content of the post.
For example, if we have,
E an X post which says: "You mean the way @globeandmail manipulates Canadians with their Liberal funded propaganda?",
E and *a claim* which says: "Major Canadian media outlets coordinate with political elites to manipulate public opinion",
we would determine the claim follows from the post since the text explicitly accuses @globeandmail of manipulating Canadians through "Liberal funded propaganda," directly aligning with the claim that major Canadian media outlets coordinate with political elites (e.g., the Liberal party) to manipulate public opinion.
#### *Generating candidate pairs*
After extracting claims, we generate a list of candidate pairs of posts and claims to run against. Because we couldn't simply run every post against every claim due to computational constraints, we then try to find relevant posts we suspect are pushing a claim. Finding relevant posts was a two-part process involving:
- 1. Finding posts containing a claim similar to instantiations and variations of each conspiratorial claim;
- 2. Initial classification using a [Qwen3-8B](https://huggingface.co/Qwen/Qwen3-8B) finetuned on this task.
For (1) we started off by manually creating a list of claims related to each conspiracy (we call these *retrieval claims*), then used [an LLM to create variations of each claim](https://arxiv.org/pdf/2503.03417), and finally found all posts containing a claim with cosine similarity >=0.5 for each *retrieval claim*. We also increased the list of candidate texts using a keyword search. For (2), we first tagged a small sample of 19,445 pairs found in (1) by leveraging [Claude Sonnet](https://www.anthropic.com/news/claude-sonnet-4-5) [4.5](https://www.anthropic.com/news/claude-sonnet-4-5) and finetuned a Qwen3-8B model on the tagged pairs. To complement our own dataset, we also trained the model on a public [stance-detection dataset](https://aclanthology.org/2022.emnlp-main.138/). Both datasets had the following class definitions:
- (1) *supporting*—text directly confirms or provides strong circumstantial evidence supporting a claim;
- (2) *refuting*—text directly contradicts the claim or undermines the claim's plausibility without explicit denial;
- (3) *other*—the text is thematically related to the claim but the claim is neutral or the model does not determine if it expresses an opinion for or against it
| Dataset | Accuracy | Fl Macro | Supporting | Refuting | Discussing | Irrelevant |
|---------------|----------|----------|------------|----------|------------|------------|
| Conspiracies | 0.770 | 0.741 | 0.807 | 0.620 | 0.744 | 0.792 |
| Stanceosaurus | 0.942 | 0.941 | 0.938 | 0.938 | 0.938 | 0.948 |
| Overall | 0.928 | 0.925 | 0.918 | 0.922 | 0.922 | 0.939 |
**Table 2: Performance results for the finetuned Qwen3-8B model on the Conspiracies and Stanceosaurus**
**dataset. We note that results shown for the supporting, refuting, discussing and irrelevant categories are their F1-score.**
(4) *irrelevant*—there is no meaningful connection between the post and the claim
In **Table 2** above, we report the performance of our finetuned model on the dataset tagged by Claude Sonnet 4.5 which we call '*Conspiracies'*, and the public stance-detection dataset which we call *'Stanceosaurus'*. Importantly, we find our finetuned model performed moderately well in identifying posts supporting conspiratorial claims in our *conspiracies* dataset. While not good enough for our liking, we note that the model proved satisfactory in at least most posts pushing conspiratorial claims.
For (1) we generated 6,566,673 of post-claim pairs (a single post can be in a pair with multiple other claims), and for (2) we identified 898,135 post-claim pairs pushing a conspiracy. Again, we note that a post can be in a pair with multiple different claims all pertaining to the same conspiracy.
#### *Final Classification*
In order to ensure maximum precision, and to stay consistent with the claims used in our survey, we leveraged the much larger [Nemotron Super 49B-1.5](https://huggingface.co/nvidia/Llama-3_3-Nemotron-Super-49B-v1_5-FP8) reasoning model and ran posts pushing a claim instantiation against the same claims that were fielded in the survey. Results shown in all sections of this brief are from this final classification.
# **Topic modelling and Named Entity Recognition**
Topic modelling was done using the [BERTopic](https://maartengr.github.io/BERTopic/index.html) library on the unprocessed text from the posts identified as supporting one of the conspiracy theories under study.
Sentence embeddings were computed using the *paraphrase-multilingual-MiniLM- L12-v2* model from SentenceTransformers.
Named entity recognition and noun phrase identification were done using the [spaCy](https://spacy.io/) library using manual entity normalization to equate synonymous names and phrases.
# **Appendix**
**Figure A1: Top 5 topics per conspiracy theory.**
Percentages give the fraction of posts discussing the topic over all posts from that conspiracy theory. The x-labels are on different scales to improve readability.
Unlike the other conspiracies, the Gender Indoctrination and Intentional Wildfire posts are categorized primarily into a single large topic. We attribute this to the relatively low number of posts for these conspiracies as well as their specific localization in time.
**Figure A2: Top 10 named entities by conspiracy theory.**
**Figure A3: Top 10 noun phrases by conspiracy theory (excluding prominent entities)**
**Table A1: Mean awareness and belief in the eight conspiracies, by demographic groups, along with their sample sizes.**
An example reading, *"Among Canadians surveyed, 349 of them belonged to the 18-34 age range, they were on average aware of 3.92, and believed on average 0.82 out of the 8 conspiracies we asked about."*
| Genaer | | | |
|----------------|-----------|--------|--|
| Group (n size) | Awareness | Belief | |
| Men (700) | 3.69 | 1.05 | |
| Women (744) | 3.28 | 0.79 | |
| | | | |
| Age | | | |
|-----------------|-----------|--------|--|
| Group (n size) | Awareness | Belief | |
| 18-34 yrs (349) | 3.92 | 0.82 | |
| 35-54 yrs (582) | 3.50 | 1.07 | |
| 55+ yrs (528) | 3.19 | 0.85 | |
| Group (n size) | Awareness | Belief |
|------------------------|-----------|--------|
| X (270) | 4.37 | 1.45 |
| Bluesky (126) | 4.34 | 0.90 |
| YouTube (816) | 3.92 | 1.05 |
| TikTok (376) | 3.84 | 0.92 |
| Instagram (678) | 3.84 | 1.00 |
| Facebook (927) | 3.59 | 0.99 |
| Low social media (161) | 2.55 | 0.61 |
| Average | Canadian | 3.49 | 0.91 |
|---------|----------|------|------|
| (1459) | | | |
# **Endnotes**
- 1 Bridgman et al., "The Causes and Consequences of Covid-19 Misperceptions: Understanding the Role of News and Social Media."
- 2 Bridgman et al., *The Canadian Information Ecosystem during the 2025 Federal Election*.
- 3 Brin et al., *Digital News Report 2025 An Overview of the Canadian Results*.
- 4 Cologna et al., "Trust in Scientists and Their Role in Society across 68 Countries."
- 5 Bridgman et al., *The Canadian Information Ecosystem during the 2025 Federal Election*.
- 6 Bridgman, *Foreign State Creators' Response to the Charlie Kirk Assassination*.
- 7 van Prooijen and Jostmann, "Belief in Conspiracy Theories"; Imhoff and Bruder, "Speaking (Un-)Truth to Power."
- 8 Hill et al., "Resonant Awakenings"; Douglas et al., "The Psychology of Conspiracy Theories."
- 9 Whitson and Galinsky, "Lacking Control Increases Illusory Pattern Perception."
- 10 Parker et al., *Old News, New Reality*.
- 11 Pehlivan et al., *The Rise of Political Influencers in Canada's Digital Information Ecosystem*.
- 12 Bridgman et al., *The Canadian Information Ecosystem during the 2025 Federal Election*.
- 13 See Questions section for the exact phrasing of each conspiracy theory.
- 14 Pehlivan et al. "Building a Media Ecosystem Observatory from Scratch: Infrastructure, Methodology, and Insights."
# **Bibliography**
Bridgman, Aengus. *Foreign State Creators' Response to the Charlie Kirk Assassination*. Media Ecosystem Observatory, 2025. https://www.cdmrn.ca/publications/foreign-statecreators-response-to-charlie-kirk-assassination.
Bridgman, Aengus, Eric Merkley, Peter John Loewen, et al. "The Causes and Consequences of Covid-19 Misperceptions: Understanding the Role of News and Social Media." *The Harvard Kennedy School Misinformation Review* 1 (2020). https://doi.org/10.37016/mr-2020-028.
- Bridgman, Aengus, Jennie Phillips, Alexei Abrahams, et al. *The Canadian Information Ecosystem during the 2025 Federal Election*. Media Ecosystem Observatory, 2025. https://meo. ca/work/the-canadianinformation-ecosystem-during-the-2025-federal-election.
- Brin, Colette, Sébastien Charlton, and Annie Marois. *Digital News Report 2025 – An Overview of the Canadian Results*. Centre d'études sur les médias, 2025. https://www.cem. ulaval.ca/publications/dnr-2025-canada-eng/.
- Cologna, Viktoria, Niels G. Mede, Sebastian Berger, et al. "Trust in Scientists and Their Role in Society across 68 Countries." *Nature Human Behaviour* 9, no. 4 (2025): 713–30. https://doi.org/10.1038/s41562-024-02090-5.
- Douglas, Karen M., Robbie M. Sutton, and Aleksandra Cichocka. "The Psychology of Conspiracy Theories." *Current Directions in Psychological Science* 26, no. 6 (2017): 538–42. https://doi.org/10.1177/0963721417718261.
- Hill, Tim, Stephen Murphy, and Robin Canniford. "Resonant Awakenings: The Social Lives of Conspiracy Theorists." *Sociology*, July 22, 2025, 00380385251344483. https://doi. org/10.1177/00380385251344483.
- Imhoff, Roland, and Martin Bruder. "Speaking (Un-)Truth to Power: Conspiracy Mentality as a Generalised Political Attitude." *European Journal of Personality* 28, no. 1 (2014): 25–43. https://doi.org/10.1002/per.1930.
- Parker, Sara, Chris Ross, Zeynep Pehlivan, and Aengus Bridgman. *Old News, New Reality: A Year of Meta's News Ban in Canada*. Media Ecosystem Observatory, 2024. https:// meo.ca/work/old-news-new-reality-a-year-of-metas-newsban-in-canada.
- Pehlivan, Zeynep, Junyan Zhu, Chris Ross, et al. *The Rise of Political Influencers in Canada's Digital Information Ecosystem*. Media Ecosystem Observatory, 2025. https://meo.ca/work/ power-shift-the-rise-of-political-influencers-in-canada.
- Prooijen, Jan-Willem van, and Nils B. Jostmann. "Belief in Conspiracy Theories: The Influence of Uncertainty and Perceived Morality." *European Journal of Social Psychology* 43, no. 1 (2013): 109–15. [https://doi.org/10.1002/ejsp.1922.](https://doi.org/10.1002/ejsp.1922)
- Pehlivan et al. "Building a Media Ecosystem Observatory from Scratch: Infrastructure, Methodology, and Insights."
# **Contributors**
## **Mika Desblancs-Patel**
Mika Desblancs-Patel is a research engineer at the Media Ecosystem Observatory. He holds an MSc in Cognitive Science from the University of Edinburgh and develops computational tools to collect and analyze large-scale social media data.
#### **Esli Chan**
Esli Chan is a senior research fellow at the Media Ecosystem Observatory. She specializes in gender issues, media and technology, and radicalization and extremism.
#### **Mathieu Lavigne, Ph.D.**
Mathieu Lavigne leads the Analytical Team at the Media Ecosystem Observatory. He completed a PhD at McGill University and a postdoctoral fellowship at Dartmouth College. He is a leading Canadian expert on misinformation, the information environment, and democratic resilience.
## **Chris Ross**
Chris Ross is a senior analyst at the Media Ecosystem Observatory. He has an MA in political science from McGill University.
#### **David Hobson**
David Hobson is a junior research engineer at the Media Ecosystem Observatory. He has an MSc in Computer Science from McGill University.
## **Ben Steel**
Ben Steel is a research engineer at the Media Ecosystem Observatory and a PhD student in Computer Science at McGill University.
#### **Helen A. Hayes**
Helen A. Hayes is the Associate Director, Policy of the Centre for Media, Technology, and Democracy.
The work of the Media Ecosystem Observatory and the Canadian Digital Media Research Network is funded in part by Heritage Canada's Digital Citizen Initiative.
