Community Notes on X: how they actually work

#Quick answer

Community Notes on X are pieces of context written by volunteer contributors and attached under a post. A note is hidden when it is first written. It only becomes public if contributors who normally rate things differently both mark it Helpful, which is why the large majority of notes never appear at all. If a note does land on your post, it stays until the ratings turn against it, and that post loses its share of X ad revenue.

Last updated: September 2026

#TL;DR

Think of Community Notes as the crowd-sourced context layer on X (formerly Twitter). Anyone who qualifies can propose a note; almost nobody's note gets shown. The gate is not a vote count, it is a disagreement test: the scoring model looks at whether the contributors calling a note helpful come from groups with a history of rating things differently. Agreement across that divide is the signal. Same-side pile-ons are not.

That design has two consequences most explainers skip. First, the odds are brutal. Second, notes arrive late enough that the post has usually already done its numbers. Below: the exact mechanism, the published data on how often it fires, what a note costs you in reach and revenue, where AI now sits in the pipeline, and what to do if one lands on you.

#How a note gets onto a post

The pipeline has four steps, and the third is the one that matters.

1. Someone flags a gap. A contributor sees a post they think is misleading or missing context. Separately, eligible accounts can hit the ••• menu on a post and request a note; enough requests push that post into the contributor queue.
2. A contributor writes the note. It has to explain what is wrong or missing and link sources. Notes are written under a pseudonymous alias, not your handle.
3. Other contributors rate it, and the bridging algorithm scores it. This is the real gate. X's scoring code, which is published in the twitter/communitynotes repo along with the full note and rating dataset, models each rater's tendencies and then asks whether a note earned Helpful ratings from raters who usually land on opposite sides. A note that only pleases one camp does not clear the bar no matter how many ratings it collects.
4. Status flips, or it does not. Every note starts at "Needs More Ratings" and is invisible to normal users. It becomes public only at "Helpful". It can also be scored "Not Helpful". Status is not permanent: a published note that loses rater support can drop back out of view.

Open data is the part worth internalizing if you care about how X works. Every note, every rating, and every status change is published daily as a public dataset. This is one of the few moderation systems anywhere you can audit yourself instead of taking a company's word for it.

#Why most notes never show up

Researchers have run that public dataset, and the results are blunt.

A four-year study covering January 2021 to January 2025 (227,702 contributors, 1,614,743 notes) found that 87.7% of proposed notes stayed stuck in "Needs More Ratings" and only 8.3% ever reached Helpful and appeared on a post. Just 13.55% of posts that received any proposed note ended up with a visible one.

Timing is the second problem. The same paper put the average time to reach Helpful at roughly 26 hours, by which point a post has typically already collected around 80% of its lifetime audience. A separate analysis in that literature found a 75.5-hour average gap between a post going up and a note attaching, after 96.7% of its reposts had already happened.

Independent work agrees on the direction. The Digital Democracy Institute of the Americas reviewed 1.76 million notes through March 2025 and found more than 90% never reached the public, with the English-language publication rate falling from 9.5% in 2023 to 4.9% in early 2025.

Read that as a system with a high bar and a slow clock, not a system that quietly notes everything. If you post a lot, your realistic odds of catching a note on any given post are small. The odds rise sharply for posts that go wide, because reach is what puts a post in front of enough contributors to get rated at all.

#What a note actually costs you

Three things happen, in descending order of how much they hurt.

Money. On October 29, 2023, X made posts corrected by a Community Note ineligible for the ad revenue share, explicitly to strip the payout incentive from sensational posts. If you are chasing the payout thresholds in X's monetization requirements, a noted post is a zero.

Credibility. The note sits under your post permanently while it holds Helpful status, in the same frame as your words, and it travels with every quote and screenshot. There is no version of your post that people see without it.

Reach, indirectly. X has said that notes reduce resharing of the noted posts. The X algorithm already weighs negative signals heavily, and a post carrying a visible correction collects more of them. Do not expect the noted post to keep climbing.

What a note is not: a strike against your account, a suspension, or a report. Notes and reporting a post on X are separate systems. Reports go to X's moderation team and can lead to enforcement; notes go to volunteers and lead to context.

#Who can contribute, and how you get writing access

The entry requirements are deliberately low but not zero. To sign up you need an account at least 6 months old, a verified phone number, and no recent violations of X's rules. Contributors operate under an alias, so your notes are not tied to your public handle.

New contributors get rating access only. You earn writing access by rating notes in ways that track how the broader contributor pool eventually scores them, which builds Rating Impact. Lose that impact through bad calls and you can lose access again. The point of the ramp is obvious once you see the incentive it blocks: without it, anyone could sign up today and note a rival's post tomorrow.

If you want to contribute usefully, the highest-value thing you can do is rate notes on topics you know well, in areas where the contributor pool is thin. The published data shows huge numbers of notes going entirely unrated, so a careful rater is scarcer than another writer.

#AI is now in the pipeline, on both ends

Two changes worth tracking if you follow X's product direction.

In July 2025, X opened an AI Note Writer API. Developers can build bots that propose notes, but those bots have to earn write access the same way people do, their notes are labeled as AI-written, and they can lose access. X framed it as "AI helping humans, humans still in charge."

In February 2026, X began testing Collaborative Notes: when someone requests a note, an AI drafts a first version and human contributors then refine and rate it together. The target is the 26-hour lag, since a fact-check that lands after the post has peaked is mostly a historical record.

Neither change moves the decision. Publication still runs through human ratings and the same bridging test, so an AI-drafted note that only convinces one side still goes nowhere.

#If a note lands on your post

You cannot delete it. What you can do, roughly in order:

• Read it properly before reacting. Notes cite sources. If it is right, the cheapest path is a follow-up post with the correction in your own words.
• Request additional review from the note's ••• menu if you think it is wrong. Notes lose Helpful status when ratings shift, and that path exists.
• Reply with better sources on the original post. You are not arguing with X, you are giving contributors something to rate against.
• Think hard before deleting. Deleting the post kills the note but also kills the engagement and looks worse than a correction to anyone who saw it. There is no undo.

#How to not get noted in the first place

None of this is exotic. Almost every note lands on a specific, avoidable shape of post:

• A number without a source. If you cite a statistic, link where it came from. Most notes are just a link you should have included.
• Stale news presented as current. Old footage or a resolved story reposted as breaking is the single most-noted genre.
• Screenshots you did not verify. Fake headlines and doctored posts spread fast and get noted fast.
• Overstated claims about a product or a person. "X is banned in Y" and "company Z admitted A" invite a correction if the real story is narrower.
• AI-generated media presented without a label. Label it, and the note has nothing to add.

That is the same discipline that makes a post worth reading anyway. The accounts that grow on X are the ones whose posts hold up when someone checks, which is a different game from the reach-at-any-cost pattern in how to go viral on X.

If your bottleneck is keeping up a daily habit of accurate, on-topic posting rather than finding things to say, that is the gap X-Autopilot is built for: it drafts on-brand posts and replies through real Chrome on your own Mac, so the routine does not depend on you being online every hour. One honesty note, because your account is the thing at stake: acting through a browser tool is a gray area under X's automation rules rather than a sanctioned API path, and no tool removes account risk. Running locally gives a smaller detection surface than cloud schedulers posting from shared IP pools, which is reduced risk, not zero.

Few parts of X can be inspected line by line. This one can. The mechanism is a disagreement test, the odds are long, the clock is slow, and the cost of catching one is real money on that post. Source your claims and the system never has anything to attach.