Lesson 4 of 6
Content Credentials & watermarks
7 min read
If a convincing fake is free to make and detectors don't work, how does anyone tell what's real? Stop inspecting the pixels — start checking the label attached to them.
Check provenance, not pixels
The durable move is [provenance](glossary://provenance): checking where a file came from and how it was made, instead of judging by looks. Two technologies carry it. [Content Credentials](glossary://content-credentials) (the C2PA standard) attach a signed "made with AI" history to a file — a nutrition label for media. Watermarks like Google's SynthID hide a detectable mark inside the pixels or audio.
Provenance is a signed label on the file — far more reliable than squinting at pixels.
Useful, but not everywhere yet
Provenance isn't magic. Not every tool adds it, credentials can be stripped by a screenshot or a re-upload, and watermarks aren't on all content. So use them as strong evidence, not a guarantee — and combine them with old-fashioned judgement: where did this come from, and does a second source agree?
A credential present is strong proof; a credential absent proves nothing — it may just have been stripped.
When an image or clip matters, look for Content Credentials and check the source — but remember a missing label doesn't mean "real", it may just mean "not tagged".
The gist
- —Provenance checks where media came from — Content Credentials (C2PA) and watermarks (SynthID).
- —A signed credential is strong proof of how a file was made.
- —Labels can be missing or stripped, so absence isn't proof of "real".
What's the most reliable way to tell if media was AI-made?
Continue in the app
Take the whole AI Media: Ownership, Deepfakes & Trust course — tracked
Get your personalized path, progress and streaks in the app — this lesson and every next one, in order.