Does Google's SynthID or Content Credentials show up when I download an image from Gemini or ChatGPT?

Workflow answer from the picture desk. Content Credentials and SynthID are two different things and you will not "see" either in a Finder info panel.

Content Credentials is the C2PA manifest: a signed block of metadata inside the file that records who made it, with what tool, and what was done to it. You need a C2PA-aware inspector to read it; a generic metadata panel will not display it. We check incoming files with one as a matter of routine, and the pattern I see is that files straight from the generator often have a manifest, and files that have been through a screenshot, a messaging app or a CMS resize often do not. Metadata is fragile.

SynthID is not metadata at all. It is in the pixels. So looking for it in a file panel will always show nothing.

To build on Marco's second point, because this is the part people find counter-intuitive. SynthID is an invisible watermark embedded in the image content itself, so there is no field to inspect. It is designed so a detector that knows the scheme can recover it, and so that ordinary transformations such as resizing, cropping or JPEG compression do not fully destroy it. You as a user cannot read it; you can only ask a detector whether it is present. That is the trade-off against C2PA: the manifest carries rich information and is easy to strip, the watermark carries almost no information except "a supported model made this" and is harder to strip. Neither is a guarantee, and "no watermark found" does not mean "not AI", it means "not a watermark this detector knows".

ok so practical question. if my client posts the image to instagram, that strips everything right? i have read that social platforms remove all the metadata on upload so after that there is no watermark and no credentials and nobody can tell. which kind of makes the whole thing pointless for anything that ends up on social

@Jules half right, which is the dangerous kind. Platforms do routinely re-encode uploads and many drop or rewrite metadata, so the C2PA manifest frequently does not survive the trip. The watermark is a different animal: it is in the pixel values, and the whole design goal is to survive the re-encoding and resizing a platform does. "Harder to remove" is not "impossible to remove", and a determined person can degrade it, but a normal Instagram upload is not that person. So after social the realistic state is: credentials probably gone, watermark probably present, and a detector from the right company can probably still find it. Not pointless. Just not visible to the reader, which is a separate problem.

@Tomas Reyes clear enough that I can repeat it on Monday. Originals in one folder, resized in another, caption on the page. And stop telling clients "there is a watermark" as though they could see it. Appreciated.

A note from the platform side on what "survives" actually looks like at scale, since Jules raised it. We inspect manifests on inbound media. A meaningful share of files that we know were made with a major generator arrive with no manifest at all, because they have been through a screenshot, a messaging app or a third-party editor before reaching us. We also see manifests that fail validation because an editor rewrote the file without updating the signed history. So the useful mental model for a marketer is: provenance metadata tells you something when it is present and valid, and tells you nothing when it is absent. Absence is the default state of most files on the open web, and that is not evidence of anything either way.