Community Notes Can Be Right After the Lie Goes Viral

In May 2023, an image appeared to show a dark plume rising beside the Pentagon. The building sat under a gray sky. Smoke climbed from an area near its white facade. The picture had the compressed, contextless look that performs well on a breaking-news feed, where a single alarming frame can outrun the work required to establish where it came from.

Accounts with paid verification shared it. News-adjacent accounts passed it along. Public reporting from the Associated Press, Reuters and Bloomberg documented how the false image circulated widely enough to produce a brief reaction in financial markets before officials said there had been no explosion. The original high-profile post vanished, but copies and screenshots remained.

That smoke plume is the useful object here. Return to it at every stage and the problem becomes plain. X could recommend the image as soon as people began replying, reposting and opening it. A Community Note, the platform’s crowdsourced label for adding context to a post, first needed someone eligible to write one.

Then other contributors had to rate it. Only after the system found sufficient agreement among people who often rate notes differently could the correction appear to ordinary users.

The image traveled as an emergency. Its correction traveled as paperwork.

The first shift belongs to the post

At publication, a post enters X without waiting for factual review. The recommendation system can place it in For You feeds based on signals such as engagement, relationships and a predicted likelihood that a user will interact. The precise weighting changes and remains partly opaque, but the basic asymmetry is public: distribution starts before Community Notes reaches a judgment.

The fake Pentagon image had several qualities that ranking systems can reward without understanding them. It concerned a globally recognizable target. It implied breaking news. It prompted users to open the image, inspect the account and search for confirmation.

Skeptical replies still counted as activity around the post. People trying to debunk it helped establish that it was worth looking at.

This is the lie’s first shift. It does not need to persuade everyone. It needs to generate enough behavior for the feed to keep testing it on larger audiences, while copies detach the claim from whichever account first posted it.

The correction faces a different standard. Community Notes contributors cannot merely declare that an image looks wrong. A useful note generally needs a concise claim, supporting sources and language that raters from different perspectives will accept. Those requirements are sensible.

They also consume the one resource a breaking feed refuses to preserve: time.

A proposed note is still invisible

X publishes Community Notes data and parts of the ranking code, allowing researchers to inspect a status history that the normal interface mostly hides. A proposed note begins in a state where it needs more ratings. During that period, the writer can see it, contributors may encounter it in the rating queue, and almost everyone viewing the post sees nothing.

That invisible interval matters more than the eventual label. A screenshot taken then records an unqualified claim. An embedded post on another site carries the same impression. A person can repost the Pentagon image to Facebook, Telegram or a group chat, where any later note attached by X has no jurisdiction.

Contributors are also responding to one post identifier. Community Notes has developed ways to attach context to matching media, including reused images, but viral misinformation rarely remains a clean duplicate. People crop screenshots, add text, change the framing or upload a screen recording. Each alteration can force the correction system to catch the claim again while the recommendation system treats the new upload as fresh material.

Public reporting around later misinformation events shows the same shape. During the false claims about Haitian immigrants in Springfield, Ohio, notes on some posts explained that a widely circulated video of a woman with a cat came from another Ohio city and did not establish the claim being made about Haitians. During Hurricane Helene in 2024, notes under some posts clarified that FEMA’s initial $750 Serious Needs Assistance payment was neither the total aid available nor a loan that had to be repaid.

Those notes addressed specific posts. The underlying stories had already split into clips, screenshots and paraphrases. A correction could win its case and still lose the feed.

Consensus has a clock built into it

Community Notes does not display a correction merely because many contributors like it. Its bridging algorithm, a ranking method designed to find agreement across people with different rating histories, looks for support from contributors who have disagreed on other notes. X presents that feature as protection against one faction controlling the labels.

It is the strongest part of the system. It is also the delay.

A note can have accurate sources and remain hidden because it lacks enough ratings from the required mix of contributors. Another can appear and later disappear as new ratings change its status. Political claims are especially difficult because the system treats cross-group consensus as evidence of usefulness, even when one group has organized its identity around refusing the relevant facts.

Peer-reviewed research using Community Notes data has repeatedly found that displayed notes can reduce subsequent engagement with misleading posts. The same research identifies the timing problem: much of a post’s sharing often happens before a note becomes visible. The intervention works on the portion of the audience that remains.

That distinction gets lost whenever X points to a correct note as proof that the system handled the post. Accuracy after display measures the quality of the label. It does not measure how many people saw the claim naked, carried it elsewhere or encountered a screenshot after the source post was deleted.

Look again at the Pentagon smoke plume. By the time a sourced note can explain that officials reported no explosion and that the image showed signs of artificial generation, the market reaction has passed, the prominent upload may be gone and the image has become a reusable file. The note is right. The event it was meant to moderate has changed form.

The platform gets speed without accepting the cost

Community Notes contributors are unpaid. They supply research, write copy, evaluate sources and rate one another’s work. X pays for the infrastructure while avoiding the larger expense, and much of the political responsibility, of maintaining a visible editorial fact-checking operation at the scale of its feed.

Meanwhile, the platform’s commercial systems remain built around attention. Eligible creators can receive revenue tied to engagement from Premium users, subject to X’s rules, while subscriptions and advertising benefit from a feed people keep checking. A Community Note does not rewind distribution or automatically reclaim money associated with attention gathered before the label appeared. X’s policies restrict harmful misinformation, but enforcement is a separate moderation decision and is not guaranteed by the presence of a note.

This arrangement works for the platform because it separates acceleration from accountability. Ranking can be fast, personalized and experimental. Correction must be careful, collectively rated and defensible. When the two systems collide, X can celebrate the visible note without publishing an equally visible account of the impressions accumulated before it arrived.

The volunteers inherit the burden of certainty. The company keeps the velocity.

A faster system would require X to accept costs it has chosen to externalize. It could reduce recommendation while a rapidly spreading factual claim attracts proposed notes, add friction before users repost disputed media, or fund trained teams to handle high-risk breaking events. Each option risks slowing legitimate information and requires appeals. Those are real tradeoffs.

They are also ordinary moderation work, which cannot be replaced by asking volunteers to reach consensus after an image is already moving.

Community Notes should remain. Its public data and viewpoint-bridging design are more inspectable than many platform moderation systems, and a well-sourced note can help users who encounter a claim later. The mistake is treating the note as the whole intervention rather than the final visible stage of a race whose starting gun belonged to the post.

The Pentagon image survives as a smoke plume in screenshots. The correction remains attached to whichever copies the system managed to catch.

Questions people ask

How does a Community Note become visible?

An eligible contributor proposes a note with context and sources. Other contributors rate whether it is helpful, and X’s bridging algorithm looks for agreement among raters with differing histories before assigning the note a status that allows public display.

Why can Community Notes take so long?

The post begins circulating immediately, while a note must be written, sourced, surfaced to raters and accepted across groups. Breaking claims also mutate into screenshots and edited copies, forcing contributors and matching systems to chase several versions of the same story.

Do

Community Notes stop misinformation from spreading?

Research indicates that visible notes can reduce later engagement with misleading posts. They cannot undo views and reposts accumulated beforehand, and a note attached on X does not follow a screenshot when it moves to another platform or private chat.

What could X do before consensus is reached?

X could temporarily limit recommendation of unusually fast-moving factual claims once credible notes are pending, while preserving access to the original post and an appeals route. That would place some cost of uncertainty on the distribution system rather than leaving all of it with unpaid contributors and users.