The Hudson Report

Meta and X are two of the largest companies currently attempting to address the spread of misinformation on social media. Meta operates Facebook, Instagram, and Threads, while X is the platform previously known as Twitter. Both companies use policies that combine community participation, content labels, reporting systems, and direct enforcement. More recently, both companies have placed a great deal of responsibility on a crowdsourced system called Community Notes. The system allows users to provide additional context underneath posts that may be false or misleading. While this approach has several benefits, its effectiveness is limited by how quickly misinformation spreads and how long it may take for contributors to approve a correction.

Meta previously relied on independent fact-checking organizations to review questionable claims on Facebook and Instagram. If a fact-checker determined that a post contained false or misleading information, Meta could attach additional information to the post and reduce its distribution. This meant that the company did not necessarily remove every misleading post, but it could make the content less likely to appear in other users’ feeds.

In January 2025, Meta announced that it would end its third-party fact-checking program in the United States and begin replacing it with Community Notes. Meta claimed that its previous system had become too complicated and resulted in too many enforcement mistakes. According to Meta’s announcement, the company wanted to focus its automatic enforcement on illegal and high-severity violations while relying more heavily on user reports for less severe violations. Meta also argued that outside fact-checkers sometimes allowed their own biases to influence what they checked and how they evaluated it.

Meta began testing Community Notes on Facebook, Instagram, and Threads in the United States in March 2025. Under this system, approved contributors can write a note underneath a post they believe requires more context. Other contributors then evaluate the note. It only becomes visible if people who have previously expressed different viewpoints agree that it is helpful. According to the Associated Press, Meta initially required notes to be fewer than 500 characters and include a link supporting the correction. Meta also used X’s open-source Community Notes system as a basis for developing its own approach.

For example, someone could post an old picture of a crowded protest and falsely claim that it depicts an event that happened recently. The photograph itself may be real, making the misinformation more difficult to recognize. A Community Notes contributor could link to the original photograph and explain its actual date and location. A similar note could be used when someone posts an AI-generated photograph as if it were real or shares a video with a caption that misrepresents what is happening.

It is an example of how misinformation does not always involve completely fabricated material. Real photographs, videos, and quotations can become misleading when someone removes them from their original context. Community Notes attempts to address this problem by adding context without completely removing the original post. Users can still see what the person posted, but they also receive information that may help them evaluate the claim.

Meta’s Community Notes system has several merits. Requiring contributors with different viewpoints to agree on a note may prevent one political group from having total control over corrections. The notes also include supporting sources, allowing users to investigate the information for themselves. Instead of seeing a general warning that simply describes something as false, users can see an explanation of what is misleading and follow a link to the evidence.

However, Meta’s new approach also removes some of the consequences that existed under the company’s earlier fact-checking system. Under the previous system, Meta could reduce the visibility of content that fact-checkers rated false. Under Community Notes, the company provides context but does not automatically reduce the distribution of every post receiving a note. This creates a situation where a misleading post can continue spreading while contributors are still deciding whether a correction should appear.

Meta’s decision to depend more heavily on user reports also places a great deal of responsibility on the same users who are being exposed to misinformation. Many users may not recognize that a post is misleading, while others may recognize a problem but decide not to report it. Ultimately, a reporting system can only address the content that users or automated systems identify. It does not necessarily address the larger structure of the platform, where controversial and emotional posts may receive more attention because they generate reactions, comments, and shares.

X uses a similar combination of Community Notes and direct enforcement policies. According to the X Help Center, contributors can write notes on any post they believe may be misleading. Other contributors rate the proposed notes, and notes become visible when enough people with different perspectives agree that they are helpful. X also publishes the Community Notes algorithm and some of its data, allowing researchers to examine how the system works.

The transparency of X’s system is one of its strongest features. Social media companies often make moderation decisions through systems that ordinary users cannot examine. By making the Community Notes algorithm open source, X gives researchers and members of the public more information about how notes become visible. The system can also place corrections underneath posts from politicians, celebrities, large organizations, and X’s own leadership. A user’s popularity does not automatically prevent a Community Note from appearing.

X also maintains policies that go beyond Community Notes. Its Authenticity policy prohibits coordinated inauthentic activity, misleading fake accounts, platform manipulation, scams, deceptive links, and certain forms of manipulated media. The policy applies to media that has been significantly altered, AI-generated depictions of real people, and authentic media that is shared with false information about its time, location, or meaning.

For example, someone could edit a video to remove important parts of a statement and then post the edited version as if it represents the speaker’s complete position. A user could also take authentic footage from one military conflict and claim that it shows a different conflict. Under the Authenticity policy, X may restrict the reach of the content, limit the account’s access to certain features, require the user to remove the post, or suspend the account in severe cases.

X also has a separate Civic Integrity policy, which addresses misinformation that may prevent or discourage participation in elections, referenda, censuses, and other civic processes. The policy prohibits false information about how, when, or where to participate. This could include a post claiming that an election has been moved to a different date, that a polling place is closed when it is open, or that voters can submit their ballots through text messages in a location where that method is not permitted.

When X determines that a post seriously violates the Civic Integrity policy, it may do more than add a Community Note. The company can place a label on the post, remove it from search results and recommendations, restrict users from sharing or responding to it, and limit its visibility to the author’s profile. This demonstrates the difference between general misinformation and information that may create an immediate risk to civic participation. While Community Notes provides additional context, direct enforcement limits the ability of a harmful claim to spread.

Still, research shows that Community Notes does not reach every misleading post. A 2024 analysis discussed by Reuters examined 283 misleading posts related to the United States election. The analysis found that 74 percent of the posts did not display an accurate Community Note to all users. Together, those posts received approximately 2.2 billion views.

The study focused on a particular sample of election-related posts, meaning that its findings cannot describe every Community Note or every type of misinformation on X. However, it provides an example of the system’s biggest limitation. A contributor may write an accurate correction, but that does not guarantee the correction will become visible. The note must receive enough ratings, and those ratings must come from contributors the system identifies as having different perspectives. By the time this process is complete, the original post may have already reached millions of users.

The requirement for agreement between people with different viewpoints is both a strength and a weakness. It may prevent partisan groups from using the system to label opinions they dislike as misinformation. At the same time, political misinformation is often the exact type of content on which contributors are least likely to agree. A note correcting the date of a photograph may receive approval quickly, while a note about an election claim may remain invisible because contributors disagree about its wording or sources.

Clearly, Community Notes can be useful in situations where a claim has a specific and verifiable answer. It works particularly well with old photographs, misidentified locations, incomplete quotations, and videos removed from their original contexts. The inclusion of supporting links also makes the notes more useful than a warning that does not explain why the content may be misleading.

However, the system is less effective when a post spreads faster than the correction process. Social media is structured around immediate reactions, while fact-checking requires time. A person can make a false claim in a few seconds, but reviewing the claim, finding reliable evidence, writing a note, and reaching agreement may take much longer. The structure of Community Notes, then, attempts to place a slower verification process within platforms designed for rapid distribution.

Meta and X could improve their policies by combining Community Notes with independent professional fact-checking. Community contributors could address straightforward claims, while professional fact-checkers could review urgent misinformation involving elections, public health, wars, disasters, and dangerous AI-generated content. This would preserve the community element without requiring volunteers to handle every complicated or time-sensitive claim.

The companies could also temporarily stop recommending rapidly spreading posts while those posts are under review. The posts would remain available, meaning the platforms would not be completely censoring them. However, their algorithms would stop actively recommending the material until a review is complete. This change would address the problem of a post receiving millions of views before a Community Note becomes visible.

Meta and X should also provide more information about whether their policies are working. They should publish the average amount of time between the creation of a misleading post and the appearance of a note. They should also report how many people see a claim before and after a correction, whether users are less likely to share posts with notes, and whether Community Notes works equally well across different countries and languages.

Ultimately, Meta and X have created systems that provide useful context, but Community Notes cannot solve misinformation by itself. The system depends on users recognizing misleading information, writing an accurate response, providing reliable evidence, and reaching an agreement before the false claim spreads too far. At the same time, Meta and X control the algorithms that determine which posts receive the most attention. Their responsibility cannot end with asking users to correct the information their platforms helped distribute. Community Notes should remain part of their misinformation policies, but it should work alongside professional review, transparent enforcement, and limits on the algorithmic promotion of content that has not yet been verified.