Methodology

How we check claims

When online verification is on, each factual claim gets one of six statuses:

• Supported
• Contradicted
• Misleading context
• Mixed or partly true
• Unverified or insufficient evidence
• Not a checkable factual claim

A status describes how well the evidence we collected supports a claim. It is not a probability that the claim is true. We never label an identifiable person as criminal, dishonest or fraudulent solely on an automated result.

Online verification is not switched on yet. Until it is, Check Am reads no sources and gives no verdicts: every completed check ends with “Online verification unavailable right now”.

Model providers

No AI model is used yet. When one is switched on, it will be Anthropic’s Claude Haiku, used through Anthropic’s API. Each report says which AI model, if any, was used for it.

Source inclusion criteria

[TO BE PUBLISHED AFTER THE RELEASE EVALUATION]

Error rates

[TO BE PUBLISHED AFTER THE RELEASE EVALUATION]

Corrections

If you think a result is wrong, use Report an issue and include your report link and any evidence. That page sets out how we triage, correct and retract reports.

Conflicts of interest

Check Am follows this rule from our Acceptable Use, Appeals and Corrections Policy (no person reviews results yet):

Reviewers declare relevant conflicts, document evidence and avoid political/financial sponsor interference. Do not allow advertisers or paying clients to override evidence standards.

Contact

Questions about this methodology: ceo@shinzii.tech. See also our Contact page.