Does a screenshot destroy content credentials?

The short answer

It destroys the content credential completely, and it may leave a SynthID watermark intact. Those two facts in one sentence are most of what people get wrong about provenance.

Why the credential cannot survive

A screenshot is not a copy. It is a new photograph of a screen, and the file it produces was created by your operating system a second ago. It has your device's colour profile, your screen's dimensions, and no history.

C2PA content credentials live in a metadata container wrapped around the pixels — a signed manifest recording what made the file and what was done to it since. Screenshotting reads the pixels and nothing else, so the manifest is not stripped so much as never carried across. The same goes for EXIF, XMP, and IPTC.

People treat this as a weakness. It is the opposite. A credential's claim is this exact file came from there, unaltered. A screenshot is an alteration, and a system that carried the signature across one would be asserting something false.

Why a watermark sometimes does survive

SynthID takes the other approach. Instead of attaching information to the file, it perturbs the content itself — small adjustments to pixel values that sit below the threshold of human vision. There is nothing beside the pixels to lose, so copying the pixels copies the mark.

A screenshot is a fairly gentle operation for that kind of signal: usually a straight pixel copy, sometimes a rescale. Google's own claim is that SynthID survives many common modifications, and a screenshot is squarely in that class. Cropping, compression, and filters degrade it gradually rather than switching it off.

So the same act destroys one signal and preserves the other. The signals are not interchangeable, and a tool that reports "no provenance found" is really reporting "none of the things I can read".

What this means when you are looking at an image

Three things follow, and all of them are worth holding onto.

• A missing credential is not evidence of anything. The overwhelming majority of honest photographs online carry none, and every screenshot of a signed original carries none either. Absence is the normal case.
• A present, valid credential is real evidence — it tells you what tool produced the file and that the bytes have not changed since. That is a narrow claim, and narrow claims are the ones that hold up.
• You cannot check the watermark yourself. SynthID detection needs Google's own model. There is a portal, it is waitlisted, and there is no public API. So the durable signal is the one no independent party can read, and the readable signal is the fragile one.

That last asymmetry is the actual state of the field in 2026, and it is not improving quickly.

The partial exception

Adobe's durable Content Credentials try to close the gap by pairing the manifest with an invisible watermark and a perceptual fingerprint, so a stripped file can be matched back to its record in a cloud database. When it works, a screenshot can be traced to its original.

It is worth knowing about and it is not a general answer. It requires the original to have been registered, it requires the matching service to be reachable, and it recovers a claim about a lookalike file rather than proving anything about the bytes in front of you. Useful, and a different kind of statement from a verified signature.

Check it yourself

Take any image with credentials, screenshot it, and put both files through the comparison tool. You will see the manifest present on one side and absent on the other, in about ten seconds, which is a faster way to understand this than any explainer.