Key Moments
How Community Notes Reduce Viral Misinformation | Keith Coleman, Jay Baxter | TED
TEDx TalksWant to know something specific about what's covered?
We've already dissected every moment. Ask and we will deliver (with timestamps).
Key Moments
Community Notes leverages user-generated corrections and an "agreement algorithm" to combat viral misinformation, demonstrating that people across the political spectrum can find common ground online. However, the system's effectiveness relies on its transparency and the inherent good faith of users, with potential vulnerabilities to sophisticated manipulation.
Key Insights
Community Notes prioritizes notes that are rated helpful by users from opposing political viewpoints, a system designed to surface "surprising agreement" and counter polarization.
Traditional fact-checking methods, often taking 2-4 days, were too slow and untrusted by the public, prompting the development of Community Notes.
A 2026 Stanford, MIT, and other university study found that reposts on X drop by approximately 50% after Community Notes are applied, indicating a significant reduction in misinformation spread.
The system has implemented defenses against manipulation, including verified phone numbers, treating users with similar rating histories as one, and monitoring for deviations from broader user consensus.
Community Notes can now be written by AI contributors via an open API, with humans providing feedback to refine AI-generated notes and improve future AI performance.
To combat the financial incentive for spreading misinformation, X now prevents posts that are noted or do not clearly label AI-generated war footage from participating in creator revenue sharing.
Harnessing 'surprising agreement' to build trust
Community Notes emerged from a desire to create a more informed world by empowering users to fact-check each other, a stark contrast to traditional top-down approaches by tech companies. The system's effectiveness hinges on its "surprising agreement" algorithm, which prioritizes notes found helpful by individuals with historically opposing viewpoints. This mechanism fosters trust across the political spectrum, as evidenced by real-world examples where detailed, context-rich notes helped debunk AI-generated imagery and debunked claims about conflicts. Unlike generic misinformation warnings, these community-vetted notes offer specific details about inaccuracies, making them more credible. The iterative process involves regular users writing notes, which are then rated for helpfulness by a diverse group before appearing on the platform, ensuring no single perspective dominates and mitigating the risk of bias. This bottom-up approach is designed to be transparent, with the algorithm's code and data publicly available for verification, eliminating the perception of corporate overreach.
The limitations of previous misinformation strategies
Before Community Notes, the internet industry, including X (formerly Twitter), experimented with various methods to combat misinformation. These included large-scale fact-checking programs, partnerships with external fact-checkers, and internal review teams. However, these efforts faced significant hurdles. Fact checks often took 2-4 days, a timeline considered an eternity in the fast-paced digital world. The scale of misinformation also overwhelmed human fact-checkers, who could only review a limited number of posts daily. Most critically, these centralized approaches suffered from a fundamental trust deficit; many users were unwilling to accept decisions about truth made by tech companies. This lack of trust spurred the development of Community Notes as a radical, crowdsourced alternative.
The 'surprising agreement' algorithm in action
The core of Community Notes lies in its algorithm, which leverages polarization as a feature rather than a bug. For any given polarizing topic, there will inevitably be individuals predisposed to disagree with a note. These individuals, when rating a note, engage in rigorous scrutiny of sources and details to support their disagreement. The algorithm capitalizes on this by only displaying notes that are found helpful by users from different political camps. This process inherently filters for accuracy, primary source usage, and neutral language. The accompanying visualization illustrates this by showing the polarizing notes in red and blue on the periphery, while the notes shown to everyone are within a green oval, signifying broad agreement across diverse perspectives. This system effectively turns partisan scrutiny into a quality control mechanism, ensuring that only the most robustly vetted information rises to the top.
No veto: upholding the principle of distributed trust
A cornerstone of Community Notes is its resolute stance against any form of override or veto. Even if posts come from highly influential figures, such as heads of state, and exert pressure through direct calls or emails to company leadership, the system's integrity remains paramount. There is no "magic button" or backdoor to remove a note. This principle was a deliberate choice, stemming from the belief that for any system to be trusted, it must be perceived as the people's opinion, not the tech company's. This unwavering commitment to a decentralized decision-making process is what, according to its creators, allows Community Notes to earn trust across the political spectrum. The system's success in influencing official statements, such as a White House retraction after a community note was posted, highlights the power vested in individual contributors.
The impact of notes on post virality and user behavior
Community Notes demonstrably curbs the spread of viral misinformation. Data shows that once a post receives a note, its viewership flattens out significantly. This reduction in views is not due to algorithmic downranking but rather organic user behavior. Informed by the added context, users naturally interact less with misleading content, resulting in fewer likes and reposts. Research from institutions like Stanford and MIT corroborates this, finding that reposts can decrease by as much as 50% after a note is applied. This indicates that users are not rigidly entrenched in their beliefs; when presented with corrective information, they are more likely to question and disengage from false claims. Encouragingly, this effect is observed across the political spectrum, suggesting a shared desire for accuracy among users. While this also leads to post authors deleting their content more frequently, which means the best notes might be seen by fewer people, the broader impact is an increase in user skepticism, acting as an inoculation against future misinformation.
Defenses against sophisticated manipulation and gaming
While Community Notes is designed to be robust, the possibility of manipulation is a constant concern. Sophisticated attacks, such as coordinated efforts by AI-driven "swarms" to manufacture consensus, are acknowledged challenges. However, the "surprising agreement" mechanism itself offers a degree of defense by requiring buy-in from diverse perspectives, thwarting simple majority rule attacks. Further defenses include requiring verified phone numbers to promote authentic human participation, treating users with highly similar rating histories as a single entity to limit the influence of sock puppet accounts, and monitoring for significant deviations between broad public consensus and potentially malicious group ratings. When incorrect notes do slip through, the system benefits from a self-correcting property: such notes attract scrutiny from a wider range of users, who then rate them as unhelpful, causing them to cease visibility.
Accelerating context with AI collaboration
Addressing the inherent speed limitations of manual fact-checking, Community Notes has integrated AI to expedite the process. Notes can now appear for new posts within approximately 20 minutes, and instantly if a URL or image matches existing flagged content. To further enhance speed and scale, an open API allows AI contributors to submit notes, with humans providing essential oversight. This collaborative model involves AI generating initial drafts, which humans then edit for accuracy, tone, and source quality. Crucially, these human corrections become training data, refining the AI's ability to generate more accurate, nuanced, and less biased notes over time. This approach, termed "reinforcement learning from community feedback," trains AI models to produce content maximally likely to be deemed helpful by a diverse set of raters, effectively enabling humans and AI to co-create better, faster contextual information.
Shifting incentives and fostering common ground
Beyond correcting misinformation, Community Notes is evolving to actively promote common ground. Incentives are being realigned, with policies now preventing noted posts and unlabelled AI-generated war footage from earning ad revenue through creator sharing programs. For repeat offenders or deliberate misuse of AI in conflict reporting, permanent suspension from revenue sharing is enforced. Looking forward, a pilot program is identifying posts and opinions liked by people from different viewpoints, not to boost them, but to signal their broad appeal. This "common knowledge engine" aims to highlight areas of agreement on topics ranging from immigration to economic policy, potentially incentivizing more constructive discourse. The ultimate vision is to create a "pro-social media" future where platforms facilitate connection and understanding, even across deep societal divides, by making common ground visible and actionable, with the potential for this open-source technology to be adopted by various platforms to foster healthier online communities.
Mentioned in This Episode
●Software & Apps
●Companies
●Organizations
●Books
●People Referenced
Impact of Community Notes on Post Engagement
Data extracted from this episode
| Metric | Change Observed | Source |
|---|---|---|
| Views after being noted | Flattens out, almost no more views | Organic user behavior |
| Reposts after notes applied | Drop by ~50% (2x reduction) | Researchers from Stanford, MIT, Udub, Paris, Luxembourg |
| Agreement with core claims after note | Decreases | Studies |
Common Questions
Community Notes allows regular users to add context to tweets. Notes are only shown if they are rated helpful by people with diverse political viewpoints, ensuring a broad consensus before being displayed, thereby reducing the spread of misleading information.
Topics
Mentioned in this video
Mentioned as having posts that have been noted, and in one case, took down a post and issued an updated statement after a Community Note critiqued it.
The organization hosting the event where this discussion between Keith Coleman, Jay Baxter, and Audrey Tong took place.
A university whose researchers have studied the impact of Community Notes, finding that reposts drop significantly after notes are applied.
Referred to as 'Udub', this university's researchers have studied the impact of Community Notes, finding that reposts drop significantly after notes are applied.
More from TED
View all 34 summaries
326 minCan Play Change the World? | Play@TED Full Event | TED, the LEGO Group and the LEGO Foundation
58 minWhy You Should Keep a List of What Makes You Laugh | Chris Duffy | TED
52 minHow to Be Smarter About the News | TED Explains the World with Ian Bremmer
44 minLove, Intimacy and Connection in the Age of AI | Bryony Cole | TED
Ask anything from this episode.
Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.
Get Started Free