X enlists AI chatbots to write Community Notes

Fact-checkers have a massive bandwidth problem. On X, misleading posts spread in seconds, while the volunteer army writing Community Notes usually only has time to flag the absolute biggest viral hits. Now, X is bringing AI bots onto the fact-checking desk to help close that gap.

X kicked off a pilot program allowing AI chatbots to generate Community Notes. Keith Coleman, X's VP of product and head of Community Notes, explained that while no specific quotas have been set, the whole point is scale. Humans simply don't want to check every single post, Coleman noted, and usually gravitate toward high-visibility content.

Bots join the fact-checking desk

Under the new pilot, these automated notes aren't limited to X's in-house Grok model. Users can create notes using Grok or plug third-party AI tools into X via an API.

Crucially, bot-written notes will not get a free pass. Every submission from an AI will be treated exactly like one written by a human. That means it must pass through the exact same consensus-based vetting system, where users who typically disagree with one another must both rate the note as helpful before it ever goes public.

According to a paper published this week by researchers working on Community Notes, the goal is to build a tandem loop. AI generates the drafts, reinforcement learning from human feedback refines them, and human raters remain the final gatekeepers.

Hallucinations meet crowdsourced consensus

Letting chatbots fact-check the internet sounds a bit like fighting fire with gasoline. Hallucinations are still an everyday issue for large language models, and models can easily invent convincing context out of thin air.

TechCrunch highlighted another subtle danger: sycophancy. If an external model is tuned to be helpful or overly agreeable rather than brutally accurate, its fact-checking attempts could be dangerously wrong.

There is also the fatigue factor. If AI tools flood the backend with thousands of half-baked drafts, human volunteers who grade notes for free might burn out fast. X says it plans to test the system for a few weeks before deciding whether to roll it out on a larger scale.

Why it matters

We think this pilot highlights the next inevitable battleground in platform moderation: using cheap synthetic labor to police synthetic noise.

Community Notes has already proven remarkably influential. Platforms like Meta, TikTok, and YouTube have raced to copy the format, with Meta even ditching its third-party fact-checking programs entirely in favor of unpaid crowd labor. If X can make bot-assisted notes work, competitors will almost certainly rush to automate their own moderation queues.

For readers, it could mean faster context on niche falsehoods, misleading political claims, and synthetic media that usually fly under the radar of human volunteers.

However, we see a serious design tension here. The magic of Community Notes relies entirely on volunteer goodwill. Flooding human raters with automated submissions risks breaking the fragile social contract that keeps the feature trustworthy in the first place. Until X proves its filters can keep AI slop out of the review queue, human raters might spend more time fact-checking the fact-checkers than cleaning up the platform.