Research Note
Synthetic Persona Verification Protocol
Verification should answer a bounded question such as "Does this account disclose that the pictured character is fictional?" or "Is this voice authorized by the person it
Synthetic Persona Verification Protocol
Verification question
Verification should answer a bounded question such as "Does this account disclose that the pictured character is fictional?" or "Is this voice authorized by the person it represents?" It should not begin with "This person looks fake."
Evidence order
| Stage | Evidence |
|---|---|
| Preserve | Canonical account URL, handle, date, post URL, visible claim, and non-invasive screenshot |
| Read | Bio, About page, pinned disclosure, captions, labels, linked site, and contact or ownership page |
| Trace | Named creator, studio, company, representative, trademark owner, and consistent official links |
| Review history | Earliest posts, disclosure changes, corrections, collaborations, and archived first-party statements |
| Inspect provenance | Available Content Credentials, file metadata, source asset, edit history, and platform production label |
| Corroborate | Independent reporting, business records, interviews, or direct confirmation appropriate to the stakes |
| State limits | Known, supported inference, unknown, and what would change the conclusion |
The C2PA 2.4 Content Credentials explainer defines a signed, tamper-evident provenance structure. It also states that provenance does not itself determine whether content is true. Credentials can be absent, removed, incomplete, or signed by an actor the reader does not trust.
YouTube's impersonation policy distinguishes synthetic-media disclosure from permission to impersonate a person or channel. A disclosure is not proof that likeness, voice, or endorsement rights were obtained.
Stop conditions
Do not contact family members, expose private registrant information, scrape personal data, run face matching, encourage dogpiling, or publish an accusation from ambiguous evidence. Escalate apparent fraud, impersonation, or nonconsensual likeness use through the platform's reporting process, the affected party, counsel, or an appropriate authority.
The defensible conclusion can be "the account does not provide enough public evidence to verify its production and representation model." Uncertainty is a result, not a failure.
Sources
Follow the evidence.
- support.google.com: 14328491support.google.com
- nature.com: s41586 023 06297 wnature.com
- solomonmg.github.io: bakshy 2015 exposuresolomonmg.github.io
- ftc.gov: federal trade commission announces updated advertising guides combat deceptive reviews endorsementsftc.gov
- FTC Disclosures 101ftc.gov
- support.google.com: 16533387support.google.com
- spec.c2pa.org: Explainerspec.c2pa.org
- youtu.be: GhR9NuCQwlMyoutu.be
- FTC Endorsement Guides questions and answersftc.gov
- support.google.com: 2801947support.google.com
- cor.inquirygroup.org: lateral reading and the nature of expertisecor.inquirygroup.org
- support.google.com: 6342839support.google.com
- newsroom.tiktok.com: how tiktok recommends videos for younewsroom.tiktok.com
- open.spotify.com: 6pC7h1LY88I1LnfvH6h4dvopen.spotify.com
- daltonanderson.ghost.io: ai influencers and the dangers of a filtered realitydaltonanderson.ghost.io