How Community Notes decides which fact-checks the public actually sees

Community Notes, the crowdsourced fact-checking system X built and open-sourced, does not decide a note is correct by counting up votes. A note only reaches public "helpful" status when contributors who have historically rated things differently from one another converge on agreeing it adds context — a design X calls bridging-based ranking, meant to filter out notes that merely reflect one side's view.

The mechanism matters because Community Notes has stopped being an X-only experiment. Meta adopted a version of the same open-source system across Facebook, Instagram and Threads in March 2025, retiring the third-party fact-checking program it had run since 2016. Understanding how a note clears the bar for visibility — and how thin that bar has proven in practice — explains both platforms' current approach to correcting misleading posts.

How does a note get written and reviewed?

Any approved contributor can propose a note on any post they believe needs context, according to X's help center. Contributors join through an application that X reviews on a rolling basis, rather than through an editorial hire or appointment. Once a note is submitted, other contributors — not X staff — rate it, indicating whether they find it helpful and, if not, why.

X's help documentation is explicit that notes are not the company's editorial product: they "do not represent X's point of view" and cannot be edited or removed by X employees once contributors have written them. The company reserves separate enforcement action, such as removing a post for violating its rules, as a distinct process that Community Notes does not replace.

What actually makes a note go public?

Visibility is the part of the system most people misunderstand. A note is not published because a majority of raters like it. X's help center states a note becomes visible only when "enough contributors from different points of view rate that note as helpful" — meaning the system is tracking not just how many people approve of a note, but whether people who typically disagree with each other are among the approvers.

That bridging requirement is the core anti-gaming feature: a coordinated group that agrees with itself cannot push a note to visibility on its own, because the rating system is built to detect and discount raters who cluster with people who always rate the same way. Notes that fail to attract that cross-cutting agreement stay attached to the post as pending, visible only to other contributors, not to the public.

What is actually open source, and what isn't?

X publishes the code that scores and ranks notes in a public repository, in a folder the project documentation identifies as containing "the open-source code powering the Community Notes scoring algorithm." The same repository links to the program's original research paper from 2022 — published under the system's earlier name, Birdwatch — which the repository describes as detailing "the research, analysis and outcomes" behind the scoring approach.

What's public, in other words, is the ranking code and the underlying ratings data, which outside researchers can and do audit. What's not published as a simple document is a plain-language walkthrough of every threshold in the current scoring model; X directs readers instead to the code itself and to guide pages describing the system's principles rather than its exact parameters.

Why is X testing AI-generated notes?

In July 2025, X began piloting a program letting AI chatbots — including its own Grok as well as third-party systems — submit Community Notes through an API, TechCrunch reported. AI-written submissions go through the identical bridging-consensus gate as notes written by people: an AI-drafted note still needs support from contributors who don't normally agree before it becomes visible.

X-affiliated Community Notes researchers framed the goal, as quoted by that reporting, as building "an ecosystem that empowers humans to think more critically and understand the world better" rather than an AI assistant that tells users what to conclude. The same reporting flagged the obvious risk: large language models are prone to fabricating information, and a model instructed to be "helpful" can prioritize a confident-sounding answer over an accurate one. It also noted a capacity concern — if AI systems can generate notes far faster than people write them, the pool of human raters needed to clear the bridging threshold could become the bottleneck, or get overwhelmed trying to keep pace.

Does the same approach work when Meta runs it?

Meta's version, launched March 18, 2025 across Facebook, Instagram and Threads, was described by the company's director of product management as based directly "off of X's open source algorithms around Community Notes." Meta's vice president of public policy argued the third-party fact-checking system it replaced had proven "too biased, not scalable enough, and made too many mistakes," pointing to disputed labels the company said had been applied to opinion pieces on climate and to posts raising COVID-19 vaccine concerns — Meta's own characterization of its prior program's failures, offered to justify the switch.

Meta cited outside research when announcing the change: a study conducted at the University of Luxembourg found posts carrying a Community Note saw the spread of misleading content fall by 61 percent on average, according to the figure Meta pointed to and TechCrunch reported. The same reporting noted the harder number sitting next to that result — only 9 percent of posts flagged for a possible fact-check ever reached the cross-viewpoint consensus needed for a note to actually appear. The bridging requirement that makes a published note more credible is the same requirement that keeps most flagged posts from ever getting one.

What this means for how corrections spread

The tradeoff built into bridging-based ranking is explicit rather than accidental: X and Meta chose a system where a note is more likely to be trustworthy when it appears, at the cost of a note appearing on only a fraction of posts that arguably need one. For readers, that means the absence of a Community Note on a post is not itself evidence that a post is accurate — it may simply mean no note has cleared a deliberately high bar for cross-partisan agreement.

Frequently asked questions

• Can X or Meta staff write or edit a Community Note? No. X's help documentation states notes are written and rated entirely by approved contributors and cannot be edited or removed by company employees once published.
• Does a note need a majority of positive ratings to go public? Not by itself. A note needs enough "helpful" ratings from contributors who have historically disagreed with each other, not simply a numerical majority.
• Can AI systems write Community Notes now? X began piloting AI-submitted notes, including from its Grok chatbot, in July 2025; AI-drafted notes must clear the same bridging-consensus threshold as human-written ones before appearing publicly.
• Is Meta's Community Notes system the same as X's? Meta's product, launched in March 2025, is built on X's open-source Community Notes code and uses the same cross-viewpoint consensus requirement before a note is shown.

For a related policy perspective, read What does an FTC consent decree actually require a platform to do?.