AI media provenance is bigger than one disclosure keyword
contains-synthetic-performer is a platform-specific disclosure value. It does not prove who created the file or how it changed. Stronger ecommerce provenance combines platform tags with creator declarations, source records, IPTC fields, and verifiable Content Credentials where available.
What XMP can and cannot prove
XMP is a flexible metadata framework, and dc:subject is a standard multi-value keyword field. Anyone with a metadata editor may be able to change unsigned XMP, so the presence of a keyword is evidence of disclosure, not cryptographic proof of origin.
Why source records still matter
- Generation history can support whether a person was entirely AI-generated.
- Creator declarations help resolve hybrid and face-replacement cases.
- File hashes connect an approval record to a specific deliverable.
- Policy-version records explain which rule was used at review time.
A practical layered model
Use the platform-required tag for disclosure, preserve descriptive metadata for operations, retain source evidence for human review, and read signed C2PA credentials when they exist. No single layer should be treated as a universal AI detector.
Evidence behind this guide
Policy guidance can change. Confirm the current Seller Central instruction before a production upload.
Common questions
Does the Amazon tag prove an image was generated by AI?
No. It is a disclosure keyword, not a signed provenance certificate.
What is C2PA?
C2PA is a technical standard for signed content provenance and edit history, commonly surfaced as Content Credentials.
Should ecommerce teams keep source files?
Yes. Source records and creator declarations are useful when pixels and ordinary metadata cannot resolve origin.
Amazon AI media tagging resources
Move from policy interpretation to writing, verification, batch processing, and provenance.