We trust evidence. Not a model.
FactCheck Surrey runs on one rule: the system never decides what's true on its own. It reads a claim and drafts a report — the verdict is bound to real evidence and signed off by a person.
The model extracts claims and drafts the report. The evidence decides the verdict — and an editor signs off before anyone sees it.
Extract the claims
Whatever you submit — a link, a screenshot, a post, or plain text — is broken into separate, checkable statements. Opinions are set aside; only factual claims are checked.
Retrieve the evidence
Each claim is searched against a curated set of sources — government records, established news, fact-check databases, primary documents. Every source carries a trust weight we set and publish openly.
This step uses no AI judgment — it's a deterministic search across known sources.Score against what was found
A verdict is assigned only by comparing the claim to the evidence actually retrieved, and it must cite the specific sources behind it. Confidence is calculated from how strong and aligned those sources are — not from the model's own certainty.
No cited evidence means the claim is marked unverifiable — never true or false.An editor reviews and publishes
Before a report goes public, an editor reviews the claims, the evidence, and the drafted verdicts. They can correct anything, and every change is logged. Only then is it published.
The claim is corroborated by reliable sources.
The claim is contradicted by reliable sources.
Technically accurate, but missing context that changes how it should be read.
No corroborating evidence was found. We don't guess.
Where AI is — and isn't — used
AI helps with language: turning messy content into clear claims, and turning evidence into a readable explanation. It does not decide the verdict, choose which sources to trust, or set the confidence. Those come from the evidence, the published source weights, and editorial review. That separation is the whole point.