An article, a news page, a post. We break it into its individual claims, check each one against primary sources, and report separately on whether the claims are true and whether the piece frames them honestly.
Every verdict cites the documents it rests on. Where the evidence does not settle a claim, we say so instead of scoring it.
Truth and framing are separate questions, so they are scored separately. Each module runs independently and reports its own evidence.
A claim is only marked accurate or unsupported when a retrievable primary document says so. The document, its publisher and the passage we relied on are listed under the claim, so you can disagree with us on the evidence rather than on trust.
Most misleading coverage is built from accurate parts. The context module compares what the piece emphasises against the fuller record: omitted findings, causal leaps, cropped quotes, base rates left out of a percentage.
Predictions, opinions and contested interpretations return “not checkable” rather than a verdict. Breaking events with thin sourcing are marked provisional and re-run as the record fills in.
Browser extension, newsroom dashboard and a REST endpoint that returns the same report as JSON.
Run your first check