Is this product photo AI? How to check listing and marketplace images
A listing photo does not show on its face how it was made. You can check the file for a signed Content Credentials record or for metadata that declares a source, and read what you find as a claim, not proof. Finding nothing does not show a photo came from a camera, because re-saving an image can strip these signals.
Can you tell from a listing photo whether it is real?
Not from the page alone. We do not judge how a picture was made by looking at it. What a file can carry is a record added by the tool that made or edited it, and that record is a claim by whoever wrote it.
So the useful question is narrower: does this file carry a provenance signal, and who stands behind it? Our checker looks for three kinds in the file you give it: a signed Content Credentials record, metadata that names a source or a tool, and the invisible watermark that Stable Diffusion's reference software adds. The file stays in your browser.
What does Google Merchant Center require for AI images in product listings?
Google's Merchant Center help page on AI-generated content says all images created using generative AI must contain metadata saying so. It tells merchants not to remove embedded metadata tags such as the IPTC DigitalSourceType property, and it names three IPTC digital source type values: TrainedAlgorithmicMedia, CompositeSynthetic and AlgorithmicMedia.
That page covers the image attributes of a Merchant Center product feed. We have not read the image rules of any other marketplace or shop platform, so this guide does not state them. Our guide on Google Merchant Center and AI images covers the Google rule. A merchant who follows it keeps a label in the file; one who ignores it may leave none. A label is only as good as the person who wrote it.
What can a checker read in a product photo?
The IPTC's Digital Source Type list defines digitalCapture as media captured from a real-life source using a digital camera or digital recording device. It defines trainedAlgorithmicMedia as digital media created algorithmically using an AI model trained on captured content, and compositeSynthetic as a mix of several elements, at least one of which is generative AI. How a mixed photo is labelled depends on the tool and the person.
| Headline | What it tells you | What it does not |
|---|---|---|
| Trusted Content Credentials declare AI use | A signed record says AI was used, and the signer is on the current C2PA Trust List. | That the record is true. |
| Content Credentials found; signer trust not established | A signed record is present and matches the image, but no list we load vouches for the signer. | Who really signed it, or whether its claims are true. |
| Metadata declares another source | Unsigned metadata names a non-generative source, such as a camera capture. | That the label is true: anyone can add or remove it. |
| No supported provenance signals found | None of the signals we read were in this file. | That the image is real or human-made. |
Can anyone add a signed camera claim to a picture?
In our test, yes, and a checker must not mistake it for a camera signature. We took a file from the Wikimedia Commons category "Images generated by Midjourney". We cannot know how it was handled before upload. We re-signed it with a certificate we made ourselves, naming "Leica Camera AG", with a record declaring a camera capture (digitalCapture) and the claim generator "Leica M11". We wrote that camera claim ourselves.
Our checker reads it as "Content Credentials found; signer trust not established". The detail under the headline says the certificate names Leica Camera AG, that it is on neither trust list, that anyone can make a certificate with any name, and that the image matches the record. We measured this on October 30, 2026, on one file; our accuracy page lists the files we made to test forged and damaged records.
The record matched the image, and the name was still not verified. The C2PA explainer says Content Credentials do not give value judgments about whether provenance data is true. Our result is about our checker and a file we made, not about Midjourney or Leica.
Why does a clean result on a listing photo mean so little?
A signal has to survive the trip from the maker to your screen. We tested one part of that trip. We drew each of the 9 files in our set that carry a Content Credentials record onto a canvas in a browser, then saved it again as a PNG and as a JPEG at quality 0.92. After the PNG re-save, 0 of 9 read as having a record. After the JPEG re-save, 0 of 9 did. All 18 outputs read "No supported provenance signals found".
The same re-save also affected the Stable Diffusion watermark in our 20 marked test files: 19 of 20 still matched after the PNG re-save, and 0 of 20 after the JPEG re-save. We measured this on October 30, 2026, in one browser at one JPEG quality; see why Content Credentials disappear.
A canvas is not a marketplace. We did not test any marketplace or shop platform, so we cannot say which ones keep or strip metadata. A listing photo you save from a web page may have lost its signals before you saw it.
What this means when you check an image
Open the checker, choose the saved listing photo, and read the headline.
If you see "Content Credentials found; signer trust not established", a signed record is present, and the line under the headline says whether the image matches it. The name in the certificate is only what the certificate says. It does not tell you who really signed it, whether a camera claim is true, or how the picture was made. Treat it as a claim from the seller or the seller's tools.
If you see "No supported provenance signals found", we found no signed record, no source label in the metadata and no Stable Diffusion watermark pattern. The line under it says this does not mean the image is real or human-made. For a saved listing photo, the signals may simply not have survived. These checks cannot determine whether an image is AI-generated.
We do not check SynthID, Meta's watermark or Adobe TrustMark. If the seller can send the original file, check that file.
Guides on this topic
- How to tell if an image is AI is the main page on this topic.
- Can Content Credentials be faked? explains self-made certificates.
- Can EXIF data be trusted? covers unsigned camera and date fields.
- How to verify a photo someone sent you is a checklist for a file you were given.
Frequently asked questions
- Does a camera claim in a listing photo's metadata prove it came from that camera?
- No. Unsigned metadata can be added or edited by anyone, and even a signed record can carry a certificate that names a company no trust list vouches for, as our Leica test file shows.
- If a checker finds nothing, is the product photo a real photograph?
- A clean result does not show that. Re-saving an image can remove these signals. Of the 32 files in our test set, 16 read "No supported provenance signals found", and we cannot say how each was handled before we got it.
- Does Google require sellers to label AI product images?
- Google's Merchant Center page says images created with generative AI must contain metadata saying so, and that merchants must not remove tags such as IPTC DigitalSourceType. That is Google's Merchant Center page; we have not read the rules of other listing sites.
- Can I check a photo straight from a marketplace page?
- You can check a saved copy, but it may have been re-encoded before you got it. We have not tested any marketplace. Ask the seller for the original file when you can.