An independent registry · Provenance Standard v1.2Verify a mark · Report misuse · EN
VERIFICATION METHODOLOGY

How we verify

Every certificate is the output of the same documented procedure. This page describes exactly what we check, against which primary source, and how a member of the public can repeat each check themselves.

Verification principle

A claim is only certified if it points to a source outside AI Trust Certify that a reasonable person could check independently. We do not ask the public to trust our judgment; we record a fact and link the record behind it. Where no such external source exists for a claim, we record the claim as disclosed rather than verified, and label it plainly.

Verifying AI disclosure

AI involvement is a self-declaration — the publisher states how AI was used, from the defined vocabulary in the Standard. We do not currently run AI-detection models, because their false-positive rates are too high to support a defensible factual claim. Instead:

  • The disclosure is recorded verbatim and permanently on the certificate.
  • It is cross-checked for internal consistency against the footage provenance (Claim 02) — e.g. a "fully synthetic" disclosure paired with an "authentic capture" footage claim is flagged for review.
  • Misrepresenting AI use, if later discovered, is grounds for revocation.
Disclosure is honest precisely because it makes no third-party accuracy claim. "The publisher declared X" is always true if they declared X — and a false declaration becomes the publisher's documented liability.

Verifying footage (C2PA)

Where the publisher supplies media with C2PA Content Credentials, we validate the manifest:

  • Confirm the cryptographic signature is intact and chains to a recognized signer.
  • Read the recorded capture / edit history and check it is consistent with the disclosed AI use.
  • Record the manifest reference (e.g. urn:c2pa:…) on the certificate so anyone can re-validate with a Content Credentials viewer.

C2PA establishes a file's chain of custody — it confirms a device signed the media and what happened to it since. It does not confirm that the camera was pointed at what the caption claims. We state that boundary openly; the credential claim is "provenance verified," never "the depicted events are true."

Verifying a physician

For any featured licensed physician we run two distinct checks against two distinct authorities:

CheckAuthorityWhat it confirms
LicensureState medical board public lookupActive license, license number, state, issue date, and any disciplinary history.
Board certificationABMS "Certification Matters" (or ABPS)Whether the physician is board-certified, and in which specialty.

A physician can be licensed but not board-certified, so these are reported separately. The certificate links to the board lookup and records a dated "verified as of." Where a free public lookup carries "not for official credentialing" language, we treat it as corroborating and confirm against the authoritative state-board record.

Re-verification cycle

Each certificate carries a 90-day re-verification date. On that cycle we re-run the licensure and credential checks and re-validate any C2PA manifest. A passed re-check extends the validity window; a failure or a lapsed window moves the certificate to Expired, and a discovered violation moves it to Revoked. All transitions are written to the certificate's verification log with a timestamp.

Known limits

We publish our limits because a credible registry states what it cannot do:

  • We verify provenance and credentials, not the truth of claims made in the content.
  • C2PA proves custody of a file, not the reality of the scene it depicts.
  • AI disclosure is self-declared; we check consistency, not intent.
  • Verification reflects the record as of the stated date; status can change between re-checks.