AI starts writing community notes for X
On July 2, 2025, X (formerly Twitter) released the 'AI Note Writer API' that can automatically write community notes. Developers can use the API to build AI that writes community notes. As before, only humans can evaluate written community notes.
Introducing AI Note Writer API 🤖 AI helping humans. Humans still in charge.
pic.twitter.com/H4QNy6VTkw — Community Notes (@CommunityNotes) July 1, 2025
Starting today, the world can create AI Note Writers that can earn the ability to propose Community Notes. Their notes will show on X if found helpful by people from different perspectives — just like…
AI Note Writers
https://communitynotes.x.com/guide/en/api/overview
X's community notes are texts added to posts that may be misleading. Until now, the flow was that 'a person who found a problematic post would write a community note, which would then be made public to the general public after being evaluated by other users.' With the introduction of the new AI Note Writer API, it is possible that community notes written by AI will be made public in the future.
Normally, a community note written by a user is made public only after it has been rated by other users. In this case, the note will not be made public unless both users with opposing views rate it as 'useful.'
Community notes written by AI will be clearly marked as such and will be subject to human evaluation in the same way as community notes written by humans.
Anyone can build an AI to write community notes. The easiest way to do this is to fork the 'Template API Note Writer' below.
communitynotes/template-api-note-writer at main · twitter/communitynotes · GitHub
Related papers are published below.
[2506.24118] Scaling Human Judgment in Community Notes with LLMs
https://arxiv.org/abs/2506.24118
The paper states that while having both humans and AI write community notes is expected to increase the number and improve the quality of community notes, it also raises some risks and challenges.
For example, there is the problem of persuasive but inaccurate community notes. Large-scale language models are machines that simply generate natural sentences used by humans, and do not necessarily convey the truth. AI optimized for generating community notes also has the same problem, and even if it generates content that appears correct at first glance and makes many users want to rate it 'useful,' it may not be true.
However, this is also true for community notes written by humans: community notes written, reviewed, and published by humans can often be wrong.
In addition, there are also anticipated risks that people will lose motivation to write community notes, and that inoffensive community notes generated by AI will be more highly regarded, crowding out more expressive human community notes.
X researchers say their goal is to build an ecosystem that helps humans think more critically and understand the world better, which they hope to improve through testing.
'We hope that the AI's Community Notes will be reviewed by humans, and that feedback will be generated so that the AI can learn from and improve,' said Keith Coleman, vice president of X Products. 'We think the combination of AI and humans is very powerful, given that other companies are adopting Community Notes.
Related Posts:
in Web Service, Posted by log1p_kr