Key takeaways
- On October 21, 2026, NewsGuard showed that X’s Community Notes (the crowdsourced fact-checking system used by Elon Musk’s social network) let 93.1% of false claims related to the US midterm elections slip through: only 13 of the 189 posts analyzed received a note.
- Once displayed, notes reduce a post’s spread by 61.2% (Nature Communications, 2026). However, because they often arrive late, the overall effect on engagement drops to −14.9%.
- The crowdsourced model is spreading across the entire ecosystem: Meta replaced professional fact-checking with notes in the United States in January 2025 and has been testing the system in Latin America since September 2026.
- For brands, the challenge is twofold: protecting their community spaces (comments, fake accounts, deepfakes) and controlling the environment in which their ads appear, since 75% of consumers view a brand less favorably when it advertises on sites that spread disinformation (IAS).
- Three levers make the difference: acting within the first few hours, building disinformation into your brand suitability framework, and combining AI with human expertise.

Crowdsourced fact-checking is now reaching its limits at scale, and this forces brands to rethink how they protect their social media spaces. Since late 2022, X has presented its Community Notes as a credible alternative to professional fact-checking: users write context notes, other users rate them, and only notes approved by contributors who usually hold differing views are displayed. The study published by NewsGuard on October 21, 2026, and covered in France by Stratégies, put this model to the test at a critical moment, the US midterm election campaign, and the verdict is clear: more than 90% of the false claims identified circulated freely, racking up millions of views. For marketing and communications leaders, this finding goes well beyond US politics. It signals a shift in responsibility for information quality, away from platforms and toward everyone who runs an online community, brands included.
What the NewsGuard study reveals: crowdsourced corrections arrive after the wave
189 misleading posts, only 13 notes: anatomy of a blind spot
The study’s central finding fits in a single line: of 189 posts spreading false election claims, 13 received a Community Note, a coverage rate of 6.9%. NewsGuard, a US organization that rates the reliability of news sources, identified allegations of voter fraud and false information about voter ID rules, which together generated millions of interactions ahead of the November 3 vote. The most telling example involves Elon Musk himself: four false claims shared by X’s owner, who has 241.5 million followers, never received a note, even though he regularly presents the system as capable of correcting anyone, himself included. This result is consistent with earlier findings: in 2025, the Spanish outlet Maldita found that only 8.3% of proposed notes are ever published, and Science Feedback reported that more than 67% of the roughly 900 disinformation tweets identified during the 2024 European elections escaped any form of moderation. The lesson is clear: crowdsourced correction covers only a fraction of misleading content, and that fraction shrinks precisely on the most divisive topics.
Real effectiveness, held back by the speed of virality
Community Notes work as soon as they are displayed; their Achilles’ heel is time. A team of European researchers (Chuai, Renault, Pröllochs and co-authors) analyzed 237,180 repost cascades on X, totaling more than 431 million reposts, in a study published in Nature Communications in 2026. Displaying a note reduces a post’s subsequent spread by 61.2% on average and increases the likelihood that its author will delete it by 94.3%. The effect weakens significantly for influential accounts and political content, and because notes often appear after the peak of virality, the overall drop in engagement is limited to 14.9%. The preprint version of this research gave a sense of scale: the half-life of reposts is 5.75 hours, and only 13.5% of helpful notes appear before that threshold. Elon Musk has also announced that users who interacted with a corrected post will receive a direct message, a feature that still has no launch date. For any exposed organization, the takeaway is operational: the value of a correction is measured in hours, and every hour lost benefits the misleading content.
The implied truth effect: when correcting a minority lends credibility to the rest
Partial coverage of false information carries a specific risk, documented by cognitive science research: it increases the credibility of content left unlabeled. In a 2020 paper published in Management Science, Gordon Pennycook, David Rand and their co-authors identified an “implied truth effect”: when only some false headlines carry a warning, people rate the other false headlines as more accurate, interpreting them as verified. Applied to the midterm elections, this mechanism suggests that the 13 annotated posts may, by contrast, have lent a veneer of legitimacy to the other 176. This dynamic also applies to brands: an inconsistently moderated comment section sends the same implicit signal to the community, which treats anything left online as validated. The rule to remember: credible moderation depends on consistency and coverage far more than on a few high-profile interventions.
Why this verification gap directly concerns brands
The crowdsourced model is becoming the norm across major platforms
The retreat from professional fact-checking extends well beyond X and is reshaping the entire social media ecosystem. On January 7, 2025, Mark Zuckerberg announced the end of the third-party fact-checking program on Facebook and Instagram in the United States, replacing it with a notes system modeled on X’s; in September 2026, Meta began testing its own version in 16 Spanish-speaking countries in Latin America. Yet Meta’s Oversight Board, a quasi-independent body funded by the company, has ruled that these notes are not sufficient to replace fact-checking. Meanwhile, the World Economic Forum ranks misinformation and disinformation as the second most severe global risk over a two-year horizon in its Global Risks Report 2026. For brands, the lesson is straightforward: vigilance over their own spaces now falls primarily to them, and it needs to be organized as a dedicated function.
Comments, fake accounts, deepfakes: disinformation is entering brand spaces
False information reaches brands through three entry points: comments under their posts, accounts that impersonate them, and synthetic content that hijacks their image or that of associated public figures. The rise of deepfakes illustrates how fast the phenomenon is accelerating: the start-up IdentifAI recorded 3,165 incidents in March 2026, compared with four in January 2020, with AI-generated video accounting for 45.6% of cases. Commercial misuse is already very real: MrBeast’s likeness was used for a fake iPhone giveaway, and Tom Hanks’s for a dental insurance scam. The public struggles to tell real from fake: an Ifop study for Cision published in March 2026, based on a sample of 2,000 French adults, recorded an average score of 5.4 out of 20 on its Anti Fake News quiz, with 89% of respondents scoring below the midpoint, and 97% of answers were wrong or incomplete when it came to identifying AI-generated faces. Every brand therefore has much to gain from mapping its points of exposure and placing them under continuous monitoring.
Brand safety: the value of an ad depends on its surroundings
An ad inherits the reputation of the content around it, and disinformation is among the costliest environments to appear in. According to Integral Ad Science (IAS), 75% of US consumers say they view brands less favorably when they advertise on sites that spread disinformation, and 51% say they are prepared to stop buying a product whose ad appears next to inappropriate content. Industry professionals share this diagnosis, but their tools remain incomplete: in the IAS-YouGov “Disinformation & Quality” study, 73% of media experts believe disinformation should be actively avoided, yet only 47% say they have clear guidelines on the matter. The Global Disinformation Index also estimates that at least $235 million in advertising spend unintentionally funds disinformation sites every year. The regulatory framework is raising transparency requirements: in December 2025, the European Commission fined X €120 million under the DSA, notably for shortcomings in its advertising repository, a decision X has appealed. Brand safety is thus emerging as a company-wide governance issue.
Three levers to take back control of information in your spaces
Act within the virality window: the first-hours rule
Response speed is the single most important factor in a correction’s effectiveness, ahead even of how well the response is written. Research by Chuai and co-authors shows that a note sharply reduces a post’s spread when it arrives early, and that its benefit diminishes as the delay grows; with a repost half-life of 5.75 hours, most of the impact is decided within the first half-day. Applied to brand spaces, this translates into a four-step protocol: continuous detection of high-risk comments and mentions; rapid classification (rumor, scam, impersonation, deepfake); a sourced public response, pinned if necessary; and reporting to the platforms through the notice mechanism provided for in Article 16 of the DSA. A brand that structures this “golden hour” turns a potential incident into a demonstration of reliability for its community.
Build disinformation into your brand suitability framework
Brand suitability, which tailors placement environments to each brand’s specific sensitivities, offers the most effective framework for tackling disinformation while preserving campaign reach. Leading advertisers are already moving in this direction: according to IAS, 45% of media experts plan to filter out entire categories of content, 43% specific topics, and 38% geographic areas where disinformation circulates heavily. DoubleVerify, working with Publicis Media and advertisers such as Kraft Heinz and Reckitt, has also shown that perceptions of the same adjacency vary depending on the ad message and audience age, with people aged 65 and over being on average 17% more likely to consider content inappropriate. The recommended approach is to formalize a written framework of sensitive topics, prioritize contextual targeting, regularly audit placements, and require from platforms the advertising transparency that the DSA makes mandatory. This framework then becomes a shared reference document for marketing, media buying and legal teams.
Combine AI and human expertise
The most robust protection combines anticipation, computing power and human judgment. Prebunking consists of familiarizing audiences with manipulation techniques before they encounter false information: in 2022, Jon Roozenbeek, Sander van der Linden and their co-authors demonstrated in Science Advances that short videos run as YouTube ads improved viewers’ ability to spot these techniques. On the moderation side, automated tools absorb volume and filter in real time, while human moderators interpret ambiguous content: irony, subverted emojis, cultural references, or weak signals of a coordinated campaign. The human role is to supervise the AI’s imperfect work and continuously improve it based on real cases encountered in the field. Brands therefore stand to gain from publishing educational content about scams that misuse their name, while relying on hybrid moderation to handle the daily flow.
Trust: brands’ new competitive advantage on social media
The NewsGuard study confirms an underlying trend: platforms are delegating verification to their communities, and that verification arrives too late to contain virality. Brands that invest in responsive moderation and a demanding brand suitability policy therefore gain a lasting asset: the trust of their audiences. This is precisely Netino’s business. As France’s leading Marketing Process Outsourcer, Netino moderates 500 million pieces of content a year for its clients, with moderation in 140 milliseconds and responses to 100% of interactions in under two hours, thanks to a combination of AI and human expertise. From detecting false information in brand spaces to securing the advertising environment, Netino helps brands turn their communities into trusted spaces, where accurate information travels faster than fake news.
Sources
- Stratégies, “Sur X, la majorité des fake news ne sont pas corrigées” [On X, most fake news goes uncorrected], October 2026. Link
- AFP / France 24, “X Community Notes ‘failed’ to flag US election falsehoods, study says”, October 22, 2026. Link
- AFP / The Manila Times, same wire story (Musk’s follower count, Meta’s tests in Latin America, Oversight Board opinion), October 22, 2026. Link
- NewsBytes, “X’s Community Notes missed 93% of false US midterm claims” (13 posts out of 189), October 22, 2026. Link
- Chuai Y. et al., “Community-based fact-checking reduces the spread of misleading posts on X”, Nature Communications, 2026. DOI: 10.1038/s41467-026-72597-0. Overview
- Chuai Y. et al., preprint version (repost half-life), arXiv:2409.08781, 2024. Link
- 24matins, “X veut corriger la désinformation avec de nouvelles alertes” [X wants to correct disinformation with new alerts] (2025 Maldita study). Link
- Science Feedback, analysis of disinformation tweets during the 2024 European elections. Link
- Pennycook G., Bear A., Collins E., Rand D., “The Implied Truth Effect”, Management Science, 66(11), 2020. Link
- Siècle Digital, “Meta met fin au fact-checking aux États-Unis” [Meta ends fact-checking in the United States], January 8, 2025. Link
- World Economic Forum, Global Risks Report 2026, January 14, 2026. Link
- Digiday, “The rise of deepfakes poses a new trust challenge for publishers” (IdentifAI report), 2026. Link
- MarqVision, “Deepfakes et protection de marque” [Deepfakes and brand protection]. Link
- The Media Leader, “Désinformation : 89 % des Français en difficulté face aux fake news” [Disinformation: 89% of French people struggle with fake news] (Ifop for Cision), March 2026. Link
- Integral Ad Science, “State of Brand Safety” (consumer perceptions). Link
- CB News, “73 % des experts média veulent éviter la désinformation et les fake news” [73% of media experts want to avoid disinformation and fake news] (IAS-YouGov, GDI). Link
- Next, “Bruxelles inflige une amende de 120 millions d’euros à X pour infraction au DSA” [Brussels fines X €120 million for DSA violation], December 5, 2025. Link
- CB News, “DSA : X fait appel d’une amende de 120 millions € infligée par l’UE” [DSA: X appeals €120 million EU fine]. Link
- The Media Leader, “Le content adjacency : les prémisses d’une stratégie de brand safety et suitability, selon DoubleVerify” [Content adjacency: the foundations of a brand safety and suitability strategy, according to DoubleVerify]. Link
- Roozenbeek J., van der Linden S. et al., “Psychological inoculation improves resilience against misinformation on social media”, Science Advances, 8(34), 2022. Link




