The technical name is C2PA: a manifest of assertions — what produced the asset, what actions were taken on it, in what order — bound to the file by a hard binding, a hash over the asset's own bytes included in what gets signed. Validation is a comparison, not a lookup: hash the file in front of you, compare it to the hash inside the signed manifest, and either they match or they do not. There is no partial credit and no degree of tampering; a hash reports one bit, same bytes or different bytes.
This is routinely confused with a watermark, and the two are genuinely different mechanisms with opposite failure modes. A watermark — particularly an invisible one woven into the pixels themselves — is built to survive exactly the operations that break a manifest: re-encoding, compression, a platform's own processing on upload. A content-credentials manifest is not designed to survive any of them, because the claim it makes is only true of one specific sequence of bytes, and every one of those operations produces a new sequence.
The practical consequence is that a manifest attached before upload is routinely gone by the time a viewer sees the file, because almost every social platform re-encodes on ingest as an ordinary part of serving video at its own bitrates and resolutions — not out of any hostility to provenance, just a pipeline doing its job. A green 'verified' tick is a claim about one exact copy at the moment it was signed, not a property that travels with the content afterwards.
In practice
- A re-encode, a screenshot or a platform's own upload pipeline all break a manifest — this is the mechanism working as designed, not a bug in a specific tool.
- Do not confuse it with an invisible watermark; a watermark is built into the pixels to survive re-encoding, a manifest is a hash that re-encoding invalidates.
- Treat 'has content credentials' as a claim about one exact file, not a property that follows the content across every copy of it.
The mistake to avoid
Treating a green verification tick as proof a piece of content was never altered anywhere. It proves that one specific copy, at the moment it was signed, matched its recorded history — a re-upload two platforms later is a different file, with no manifest left to check at all.
Go deeper
Why C2PA manifests die on upload
A C2PA manifest is a hash over exact bytes, so re-encodes and screenshots kill it by design. How to find the hop in your pipeline where provenance is lost.
Related terms
Watermark
A watermark is a mark applied to generated media to identify where it came from — either visible on the picture or embedded invisibly in the file.
AI content label
An AI content label is the platform-applied or creator-applied tag marking a post as made or meaningfully altered with AI — a visible badge, distinct from the legal disclosure obligations that exist independently of any single platform's UI.
Synthetic media disclosure
Synthetic media disclosure is the broader legal obligation to identify AI-generated or altered content as such — increasingly a binding requirement rather than a platform preference, with the EU AI Act's Article 50 as the clearest current example.
Hook rate
Hook rate is the share of people shown a video who are still watching a few seconds in — the number that grades the opening, not the edit behind it.
Retention curve
A retention curve shows how many of the people who started a video are still watching at each second of it, and watch time is the area under that curve.
The all-in-one AI studio for creators. 60+ models for video, image, voice, music and lipsync in a single app.