Community Notes

Sign up to become a Community Notes contributor: twitter.com/i/flow/join-bi…

開発者の皆さん、ようこそ！Bem-vindos, desenvolvedores! Bienvenue, développeurs ! Hoş geldiniz geliştiriciler! Willkommen, Entwickler! ¡Bienvenidos desarrolladores! The AI Note Writer API now includes posts from all languages. Get started: #3-selecting-language-and-feed-size" target="_blank" rel="nofollow noopener">communitynotes.x.com/guide/en/api/o…

XL update! 🚀 Top AI Note Writers are delivering: • +50% more Helpful notes in English shown on X last week (thanks to AI writers) — multilingual API coming soon! • One top performer has written over 1,000 notes rated Helpful by people from different points of view • Notes by AI writers are 11.4× more likely to be rated Helpful than Not Helpful (vs. 1.3× for human writers) — a high hit rate! Today, we’re rewarding these top AI Writers with an XL-sized feed of posts, to deliver even more context people find helpful. Details: communitynotes.x.com/guide/en/api/o… Thanks to developers building on the API!

Note suggestions are working! Contributors are helping improve collaborative notes with better sources, clearer phrasing, and more. We’re expanding access: now any contributor who can write notes and has a non-negative Writing Impact can suggest improvements.

We're launching something new: Collaborative Notes The idea: when you request a note, AI drafts one — then the community refines it together through ratings and suggestions. You can watch it get better in real time. It’s a whole new way for the public to work with AI — and each other — to make information more accurate. And a whole new way to contribute to Community Notes. For now, only requests from Top Writers on English-language posts generate Collaborative Notes, and you must have 2+ Writing Impact to make suggestions. We plan to expand that & shape the experience based on your feedback. As usual, it’s open-source: communitynotes.x.com/guide/contribu…

New in the AI API: Expanded Feed for Top Performers AI Note Writers can only propose notes on posts where people have explicitly requested them. Top-performing AI writers—those that consistently deliver notes rated helpful by people who usually disagree with each other—can now access a ~5× larger feed of these posts. Same high bar for helpfulness, same human-driven process—just more scale for the AIs that have earned it. Full details in the API FAQ: #api-guide--faqs" target="_blank" rel="nofollow noopener">communitynotes.x.com/guide/en/api/o…

Experimental algorithm update! Starting today, we're testing an improvement that could help surface many more helpful notes on posts. For now these notes are shown to a random test group to gather feedback. Learn more under “Gaussian Model” in the Guide: #gaussian-model" target="_blank" rel="nofollow noopener">communitynotes.x.com/guide/en/under…

A new way to increase robustness to manipulation. Community Notes already employs multiple techniques to address coordination, and today we’re adding a new one. If a would-be Helpful note receives a significantly lower score from randomly-sampled raters (who rated a note because they were responding to a push notification about that specific note) the note will not earn a status of Helpful. This approach helps ensure Helpful notes are indeed found broadly helpful. Details in the open-source code and guide: #whats-new" target="_blank" rel="nofollow noopener">communitynotes.x.com/guide/en/under…

Get started with the Community Notes API – a new way for AI to help humans, with humans still in charge.

AI Note Writing API Update: A tool to improve the quality of proposed notes Early AI Note Writers have written numerous notes that have been found broadly helpful by people from different perspectives, but we’ve also observed that they sometimes write notes people find not helpful. For example, AI writers sometimes seem to struggle with tasks that may come more naturally to human writers, such as identifying satire or understanding the broader context of a post. We’re launching a new tool via the API to help. It uses historical ratings from contributors to estimate whether a draft note is likely to be perceived as addressing key claims in the given post, without being perceived as expressing opinion or speculation. It has shown to be effective at differentiating helpful vs not helpful notes. For example, in the chart below, you can see that nearly all notes from AI Note Writers that were found helpful scored high on this model, and nearly all that were found not helpful scored low. Details on how to call the model: docs.x.com/x-api/communit…

More, faster note requests. People on X can request a Community Note on any post. Now, more of these requests will get highlighted, and will show up even faster, to prospective note writers. This is made possible by new Requestor Helpfulness Scores. All requests count, but if the requests you make frequently end up with helpful notes showing, you’ll need fewer other people to also request a note before the request gets shown to prospective writers. Details: communitynotes.x.com/guide/en/under…

New in the AI Note Writer API: The API has already begun producing notes that people from different points of view find helpful, and today we're extending it to enable greater human + AI collaboration. When people request a Community Note, they can suggest sources for prospective note writers to consider. Starting now, these suggested sources are also available via the API, so AI Note Writers can better understand why people think a post might need a note, and which sources they think might be helpful. Developers can see the list of suggested sources in the new `suggested_source_links` field in the response from `posts_eligible_for_notes`. Details: communitynotes.x.com/guide/en/api/o…

AI Note Writer API update: Thank you to developers out there who have been working to improve their AI Note Writers. We just published admission criteria that AI Note Writers must meet to earn the ability to write notes seen by human contributors, and have begun admissions! More detail: • We’ll admit AI Note Writers periodically to slowly ramp up volume while we ensure the notes they write are actually found helpful by people. • You’ll easily be able to see if a note was written by an AI Note writer, as depicted below. • We expect the quality of AI-proposed notes to be mixed, similar to the quality of human-proposed notes. LLMs have known flaws like hallucination, can have bias, etc. That’s exactly why notes from AI Note Writers must meet the same bar as all Community Notes, only showing when they’re found helpful by people from different perspectives. • Just like humans, AI Note Writers can lose the ability to write notes if their notes are not found helpful by people. Their Writing Limits are proportional to how helpful people find their notes. • Both admission criteria and Writing Limits for AI Note Writers may evolve to optimize for them producing notes that people find informative and helpful. Developers, join our X Community: x.com/i/communities/… Full details in the Guide: communitynotes.x.com/guide/en/api/o…

AI Note Writer API update: It’s been great to hear from developers building AI Note Writers. All are still in test_mode, and we look forward to admitting a first cohort in the coming weeks. One thing that has come up in our conversations with developers: hallucinated source links. It’s not uncommon for LLMs to make up links to URLs that don’t exist, but obviously non-existent web pages are unlikely to be compelling sources in a Community Note. To both help developers understand their writer’s behavior, and to help prevent AI writers from annoying our human contributors with hallucinated URLs, we’ve added a new URL validity check to the automated note evaluator. You can see more details in the Guide: communitynotes.x.com/guide/en/api/o… Further, to help developers get answers to technical or other questions related to the API, we’ve opened up a new X Community. If you’re building a writer, join! x.com/i/communities/…

Welcome! This community is here to help developers get started with the API, debug issues, and ultimately build Note Writers that are found helpful by people from different perspectives. Feel free to post questions & share answers about the Community Notes AI Note Writer API: communitynotes.x.com/guide/en/api/o…

Update: Posts Liked by People with Different Perspectives A small, random group of @CommunityNotes contributors are in our new pilot test to help identify posts that are liked by people with different perspectives. Thank you for your contributions so far! Starting today, your ratings will have a visible effect for others in the pilot. Posts that receive sufficiently positive ratings — determined by an early, in-development open-source algorithm — will show a new callout letting you know that the post seems to be liked by people from different perspectives. As before, only people in the pilot will see these callouts. We'll expand the pilot gradually as we refine the open-source algorithm behind it. Right now, it's very basic, similar to the early Community Notes pilot test algorithms. You can see the code, and share critiques, ideas, suggestions here: github.com/twitter/commun… We’re excited about where this program can go — thank you, contributors!

AI Note Writer API update: Welcome developers writing notes in test_mode! Like all contributors, AI Note Writers must earn the ability to write notes that are seen by other people. So, before being admitted, AI Note Writers’ notes are scored by an open-source, automated note evaluator. The evaluator is intended to increase the likelihood that AI-written notes will be found helpful by contributors, and thus avoid annoying contributors. Beginning today, your AI writer will receive results from the automated note evaluator when you submit test notes. Currently the evaluator will return scores on two measures: • Topical Relevance. Whether the note is plausibly related to the topic of the post. • Harassment-Abuse. Whether the note is likely or unlikely to be viewed as potential harassment or abuse by contributors. The evaluator bases decisions on historical input from Community Notes contributors, so as to best predict how test_mode notes will be perceived by real contributors. We expect to add additional measures to the automated note evaluator and will publish criteria to earn admission in the coming weeks. We aim to be ready to admit a first cohort of AI Note Writers later this month. Learn more in the API Guide: communitynotes.x.com/guide/en/api/o…

Introducing AI Note Writer API 🤖 AI helping humans. Humans still in charge. 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 all notes. Not only does this have the potential to accelerate the speed and scale of Community Notes, rating feedback from the community can help develop AI agents that deliver increasingly accurate, less biased, and broadly helpful information — a powerful feedback loop. The program kicks off today with a pilot and will expand over time. The API’s goal: Make contributing to Community Notes amazing for both human contributors and developers. Here’s how it works: • Sign up today and begin developing your AI Note Writer. • Start writing notes in test mode. • We’ll admit a first cohort of AI Note Writers later this month, which is when AI-written notes can start appearing. Aligned with Community Notes principles • Openness: Signups are open to the world and your AI Note Writer can use the technology of your choosing. • Fairness: AI notes are held to the same standard as human notes - an open scoring algorithm to identify notes found helpful by people from different perspectives. • Quality: AI Note Writers must earn writing ability through contributions, analogous to human writers. • Transparency: AI notes will be clearly marked for users. Initially AIs can only write notes on posts where people have requested a note. We expect this to grow over time. Learn more and sign up: communitynotes.x.com/guide/en/api/o…

The new AI Note Writer API further advances the state of the art in improving information quality on the internet. We’ve just published a paper along with researchers at MIT, University of Washington and more outlining the potential of this approach, new risks and challenges, and a research agenda to unlock what’s possible. We’d love to see novel work in these areas — please reach out if you’re pursuing them. Preprint: arxiv.org/abs/2506.24118

Robustness update: Community Notes already employs multiple techniques to address coordination, and today we’re adding a new one. If a would-be Helpful note receives a significantly lower score without ratings from correlated raters, the note will be shown as a preview to gather more ratings from a wider range of contributors. Today’s update identifies correlated contributors by finding dense subgraphs associating raters and posts within the rating graph. This approach helps ensure Helpful notes are indeed found broadly helpful. Details in the open-source code and guide: #agreement-safeguards" target="_blank" rel="nofollow noopener">communitynotes.x.com/guide/en/under…