How Veratify checks a claim

Veratify publishes fact-checks of widely shared claims on social media. Each check names the claim, gives one verdict, explains the evidence, and links every source we relied on, so you can repeat the work yourself.

Where the claims come from

We do not pick claims by who said them. Our candidates come from X's Community Notes program: posts that X users flagged as possibly misleading and asked to have a note added. Our first published checks (August 2026) came from Community Notes' public data on posts where contributors flagged an error but never reached agreement; since September 2026 our source has been X's list of posts on which users requested a note.

Since 31 August 2026, before any check runs, an automated screen sets aside posts that are not about a consequential public matter (for example sports or entertainment trivia) so that our attention goes to claims where a false belief can cause real harm: policy, health, elections, the economy, safety and scams.

How a check works

The claim is isolated. We extract the specific factual assertion from the post. The checking system sees the post itself, never the Community Notes contributors' opinions about it.
Sources are searched live. Priority goes to primary sources — government data, regulators, court records, peer-reviewed research — and to established news organizations. A person's own denial of an accusation is not treated as evidence.
No sources, no verdict. A check that cannot point to its sources is not published.
Health claims meet a stricter bar. A health or medical check is published only if its evidence includes a primary medical source: a public-health agency, a regulator, or peer-reviewed literature — not only news coverage of one.
A human editor decides. Every published check is reviewed and approved by Veratify's editorial board before it appears here. Candidates the board declines are recorded with a reason from a fixed list: low confidence, duplicate, not objectively verifiable, stale, legal risk, or out of scope. The board can correct an automated verdict before publication.

The checking itself is done by an automated system that uses AI models with live web search. It is built to decline — to say it cannot establish a claim — rather than guess, and checks it is not confident in are not sent to the board at all.

The verdicts

Every published check ends in exactly one of:

Credible — supported by authoritative, independent sources.
Partially true — a real core with something meaningful wrong or missing; the check says what.
Contested — authoritative sources genuinely disagree.
Likely false — contradicted by authoritative sources.
Insufficient data — the evidence available does not settle it either way, and we say so rather than guess.

We judge the claim as it was made. If a post attaches a real event to a false cause or the wrong person, the verdict follows what the post asserted, and the check states plainly what is actually true.

Because our candidates are posts that people flagged as possibly misleading, most of our verdicts are "Likely false" or "Partially true." That reflects where the candidates come from, not a preference: when the flag was wrong, we publish "Credible."

The running tally

Every published check is tagged by topic, verdict and the kind of person or organization that made the claim. The running totals are published on this site and update automatically, so anyone can see what we check and how it comes out.

Nonpartisanship

We apply the same standard of evidence to equivalent claims regardless of who made them. Veratify is not affiliated with any political party, candidate or advocacy organization. We do not aim for an equal number of checks on each political side: which claims we check follows where flagged, consequential claims are circulating, and the running tally makes that visible.

Corrections

When we get something wrong, we correct the check in place and record the correction, with its date and the earlier verdict, on the check itself. See our corrections policy.