An image's credentials show its history, not whether it is real

Start here

A reader looking at a striking photo or video has one more question available than 'does this look real': can I check where this came from. Two overlapping systems attempt to answer that, a provenance standard called Content Credentials and a watermarking system from Google DeepMind called SynthID. Neither solves the underlying problem alone, and the documents say so plainly. The task is understanding what each check can and cannot tell a reader, as the tools stand when retrieved on 16 September 2026.

What the documents say

Content Credentials, built on a standard from the Coalition for Content Provenance and Authenticity, work like a 'nutrition label' attached to a file: signed metadata showing who or what created the content, what tools were used, and what edits followed, tamper-evident because later changes break the signature. The standard's documentation states directly that credentials 'aren't intended to prescriptively indicate whether a piece of content is real'; they describe origin and history, not truth. SynthID takes a different approach, embedding an imperceptible watermark into images, video, audio, and text generated by Google's own AI models, checkable through Gemini or a separate detector. Google's own verification guide names a key limit: SynthID confirms only content made by Google's models, and a negative result does not mean content is authentic, since it 'could have been created by other AI systems' carrying no such watermark.

Check this

The concrete check is to look for a Content Credentials icon on an image from a participating platform or camera, and use a verification tool to view its attached history, rather than judging authenticity by eye. Where a Google-AI origin is suspected, Gemini's built-in check or the support page's detector can confirm a SynthID watermark. Absence of either signal proves nothing: an image with no Content Credentials may simply come from a source, screenshot, or export path that never attached them, and no SynthID watermark only rules out one company's models, not every generator that exists.

What holds and what fails

Both systems hold when they detect a positive signal: a valid Content Credentials chain or a confirmed SynthID watermark is strong, specific evidence about origin. Both fail as a general authenticity test, since neither covers every image in circulation, credentials can be stripped by conversion, and absence of a signal is uninformative rather than reassuring. Treating 'no watermark found' as proof an image is genuine is the most common misreading of what these tools state about themselves, and the opposite of what their documentation says.

• Look for a Content Credentials icon before trusting a striking image's origin story.
• Remember that no watermark detected does not mean an image is authentic.
• Use a verification tool's history view, not visual judgment alone, when the stakes matter.

Provenance tools are becoming more available, but each documents itself as a partial signal rather than a verdict, which makes the habit of checking more valuable than any single tool's result.