Community Notes was built on a simple promise: instead of X deciding what counts as misinformation, its users would. A new preprint from a group of X engineers, analyzed this month by Indicator, shows how far that promise has drifted. The notes readers see most often are now written largely by X itself, through an open-source AI notewriter called the Community Writer, which the engineers describe as the program's largest AI API client.
The numbers are striking. In September, the six accounts powered by the Community Writer accounted for more than 90 percent of all notes contributed by AI writers. More than that, those six accounts wrote 55 percent of all notes rated "Helpful" — the notes actually shown to users beneath posts. According to the preprint, the six bots generated about 240,000 draft notes in June alone, of which roughly 3,600 were rated helpful.
That output is not obviously better than everyone else's. Counting deleted notes, about 9 percent of the X-run writers' notes are rated helpful, in line with other top-performing AI bots and below the most prolific human writers. What X has is scale and access. AI writers start with a quota of 10 notes a day and can earn their way up to 500 a day with a strong helpfulness record, but 88 percent of all notes, whoever writes them, sit stuck in "Needs More Ratings." X runs its own models on its own systems, can draft tens of thousands of notes at no API cost, and its writers arrive first with a proposed note on 90 percent of posts that another AI bot also notes.
One safeguard remains intact: people still decide what readers see. A note appears only when raters who have disagreed with each other in the past agree it is helpful, so the community filter still applies even when the writer is a machine. X has also published the Community Writer's code and shares program data, a level of transparency that Indicator notes Meta and TikTok have not matched with their own crowdsourced fact-checking features.
The concern is about who, or what, does the writing. Indicator's analysis found that X's AI writers mention elections, Israel and Gaza, vaccines, or immigration in about two-thirds as many notes as other AI writers do, and cover political topics noticeably less in a manual review of 150 sampled helpful notes. A system that rewards cross-partisan helpfulness can quietly steer writers, human or AI, away from the contested topics where context matters most.
For everyday users, the takeaway is practical. When a Community Note appears under a post, there is now a good chance an X-operated bot drafted it, and a diverse group of human raters approved it. The raters are still the crowd. The writers, increasingly, are not — and that shift is worth watching as the program grows.