AI Claim Decoder
Paste a sentence from an AI product page, news story, policy document, paper summary or social post. The tool does not decide whether the claim is true. It asks: what kind of evidence would this wording require?
Privacy: this page runs entirely in your browser. It has no analytics, account, server upload, external scripts, or AI API. Reloading discards what you typed.
Evidence is claim-shaped
Capability
“Can do X” needs task-specific tests, edge cases and a clear population of tasks.
Reliability
“Works reliably” needs repeated measurements, failure rates and deployment conditions, not one demo.
Architecture
“Runs locally” or “uses retrieval” can often be checked from technical design, logs or code, but architecture alone does not prove outcomes.
Privacy / security
Needs data-flow, retention, access-control and threat-model evidence. “We care about privacy” is not the same claim as “data never leaves your device.”
Welfare / mind claims
Terms like conscious, sentient, understands or feels carry philosophical and scientific assumptions. No simple benchmark turns them into settled facts.
Prediction
“Will replace,” “will transform,” or “will never fail” is a forecast. Ask for timeframe, conditions and calibration rather than treating confidence as evidence.