Chapter 12 · Field guide
How Does Community Notes on X Work?
Short answer
Community Notes lets eligible X users write and rate context notes on posts, and a note only appears publicly when contributors who usually disagree with each other both rate it helpful, as measured by an open-source bridging algorithm.
Key takeaways
- Anyone can read notes; contributors need an account at least six months old, a verified phone number from a trusted carrier and no recent rule violations, per X's guide as of November 2026.
- A note needs at least five ratings, and its helpfulness score must reach 0.40 with support from raters on both sides of the model's main viewpoint split.
- Notes, ratings and the scoring code are public, so researchers can check how the system behaves.
- Most proposed notes never show, and notes often arrive after a post has already spread. Treat a missing note as no information.
You open a viral post and see a box underneath it: "Readers added context." That box is a Community Note. It looks like a verdict from the platform, but it comes from a crowd of volunteers and a public algorithm with some unusual rules. Knowing those rules helps you judge how much to trust a note, and how little to read into the lack of one.
What is Community Notes?
Community Notes is X's crowd-sourced system in which contributors write context notes on posts and other contributors rate them, with an algorithm deciding which notes become public.
The program started in 2021 under the name Birdwatch and was later renamed. X describes the goal as adding helpful context to potentially misleading posts. Everyone can see notes that reach public status. Only enrolled contributors can write or rate them.
Notes typically do one of a few things: give a missing date, link the original study or video, point out that an image is edited or old, or quote the source the post misrepresents.
Who can contribute notes?
According to X's Community Notes guide as of November 2026, contributors need an account that is at least six months old, a verified phone number from a trusted carrier and no recent notice of rule violations.
The phone number must not be tied to another Community Notes account, which is meant to make it harder to run several contributor accounts. New contributors start by rating notes. The ability to write notes comes after their ratings show they can judge helpfulness in line with other raters. X says these criteria may change as the program evolves, so check the guide for the current version.
How does a note get shown?
A note is shown when its helpfulness score reaches 0.40 and the support comes from raters who sit on both sides of the model's main viewpoint split, not simply from a majority.
This is the bridging idea, and it is the part most people get wrong. The algorithm is a form of matrix factorization. It learns, from past ratings, a viewpoint position for each rater and each note. A note's score is the part of its helpfulness that the viewpoint does not explain. In plain terms, a note rated helpful only by one side scores low, even if that side is large.
The published documentation lists several thresholds:
| Rule | Value in the public documentation |
|---|---|
| Ratings before a note can leave "Needs More Ratings" | At least 5 |
| Helpfulness score (intercept) to be shown | 0.40 or higher |
| Maximum lean toward one viewpoint | Factor magnitude under 0.50 |
| Raters needed from each side of the split | At least 5 positive-factor and 5 negative-factor raters |
| Score at which a shown note loses status | Below 0.39 |
Rating options map to numbers too: "Yes" counts as 1, "Somewhat" as 0.5 and "No" as 0. The code that does all this sits in a public repository, and X publishes the notes and ratings data so outsiders can rerun the scoring.
What are the limits of Community Notes?
Most proposed notes never appear, and the ones that do often arrive after a post's fastest spread.
The bridging requirement is a strength and a bottleneck. On contested topics, raters split along familiar lines and few notes clear the bar. A 2025 analysis by the Spanish fact-checker Maldita reported that only a small share of proposed notes became visible. Academic work published in 2026 found that notes reduce the spread of posts once attached, but often come too late to catch the early, most viral stage.
So keep three things in mind:
- No note is not a clean bill of health. It may mean raters disagreed, or nobody wrote one yet.
- A note is a claim with sources. Click them. Notes are usually good, and occasionally wrong.
- Notes can be removed. If later ratings drop the score below 0.39, the note disappears from public view.
How do AI note writers fit in?
X has opened an API that lets developers build AI note writers, and in 2026 it tested AI-drafted notes that contributors can request and refine, but human ratings still decide what is shown.
Reporting in 2026 described a pilot where contributors could ask for an AI-written first draft, with Grok as the writer, and then rate or improve it. X also announced in July 2026 that people who engaged with a post may get a message when a note is later shown on it, though coverage disagreed on how far that feature had rolled out. Treat both as features that can change.
How should a skeptic use Community Notes?
Use a note as a lead to follow, not a verdict to adopt, and do your own quick check when a post matters to you.
Read the note's sources the way you would read any source: open them, see whether they say what the note claims, and check who published them. The source credibility checklist works well here. If you want to contribute yourself, the guide's advice matches good practice anywhere: cite sources, stick to verifiable facts, and write so that someone who disagrees with you would still find the note useful. That last rule is the whole design of the system, and it is a decent rule for replies too.
Frequently asked questions
Does a Community Note mean a post is false?
No. A note adds context that raters from different perspectives found helpful. That can mean a correction, a missing date, or a link to the original source. Read the note and its sources the way you would read any claim.
Does X staff write or choose the notes?
Notes are written and rated by contributors, and the algorithm decides which ones show. X sets the rules, runs the program and builds the scoring code, which it publishes.
If a viral post has no note, is it accurate?
Not necessarily. Many notes are proposed but never reach the threshold, and notes can take hours to appear. A missing note tells you nothing about accuracy.
Can AI write Community Notes?
X has piloted AI note writers that submit draft notes, which are then rated by human contributors like any other note. In 2026 X also tested AI-drafted notes that contributors can request and edit. Human ratings still decide what shows.
Sources
- Community Notes Guide (X)
- Community Notes open-source repository and documentation (GitHub)
- Community Notes documentation: how notes are ranked
- Community Notes documentation: signing up
Examples in this chapter are invented for teaching. Facts about platforms were checked in November 2026 and can change. See how we research.