X will let AI chatbots generate their own Community Notes, which are user-submitted comments attached to some posts that offer additional context. The pilot program begins Tuesday, July 1.
X hopes this will help Community Notes scale, Keith Coleman, X’s vp of product and head of Community Notes, told ADWEEK. X has not established specific targets for growing the new AI-generated Community Notes program.
“Our focus has always been on increasing the number of notes getting out there,” he said. “And we think that AI could be a potentially good way to do this. Humans don’t want to check every single post on X—they tend to check the high visibility stuff. But machines could potentially write notes on far more content.”
Users can build their own AI note writers using LLMs like X’s proprietary Grok or another system like OpenAI’s ChatGPT, and connect them to X using an API.
For X, public access is a critical principle. “We want all of humanity to be able to participate. And we want people to develop the best possible technology to solve this problem—whatever that is,” said Coleman.
Notes submitted by AI systems will need to adhere to the same rules as human-written notes. They will undergo ratings by human users to help validate their accuracy, and, in order to be displayed to all users, they will need to be found “helpful” by users with historically differing perspectives. AI-generated Community Notes will also be subject to X’s scoring mechanisms, which attempt to mitigate inaccuracies and misinformation.
“These AIs, like humans, are going to be able to propose notes, but they still get vetted through the same process as all Community Notes,” Coleman said.
The program is not intended to supplant human contributions, Coleman said, arguing that both human- and AI-written notes “are going to be very additive.”
AI-generated Notes will appear in a few weeks
Although the pilot launches today, everyday X users will not begin seeing AI-generated Community Notes in their feed for a few weeks, when an initial cohort of testers will begin publishing them.
X is optimistic that human feedback to AI-generated Community Notes will feed into AI systems’ training and cyclically improve the accuracy of their notes.
“Feedback from the community could actually help AI models get better at writing more fair and more accurate context,” said Coleman. “Right now, they have whatever training data they have, but this is going to give them feedback from people of many different perspectives, and they can use that to fundamentally improve what they write in the first place.”
The launch coincides with a research paper published Monday by X’s Community Notes leaders in collaboration with researchers at MIT, Stanford, Harvard, and the University of Washington that advocates for adoption of “an open ecosystem where both humans and LLMs can write notes, and the decision of which notes are helpful enough to show remains in the hands of humans.” Researchers suggest that by using both human and AI contributors, Community Notes can expand its reach and speed without diminishing trust.