How to check Content Credentials on an image, step by step, and what each result means
Open our checker in your browser, choose the image, and read the headline. The check runs on your device. If a headline says a signed record was found, it tells you who the certificate names and whether the image still matches the record. It does not tell you that what the record claims is true.
What do I need before I start?
You need the image file itself, as received. Content Credentials are a signed record, usually stored inside the file. They can be removed, and the C2PA explainer answers the question "Can the provenance metadata be removed?" with "Yes it can." A screenshot or a re-save is a new file and may carry no record at all.
Our checker reads JPEG, PNG, WebP, AVIF, HEIC, HEIF and GIF images, one at a time, up to 50 MB. A file in another format, such as an SVG or a PDF, gets the headline "Unable to check this file".
How do I check Content Credentials with the New Watermark checker?
- Open the New Watermark checker.
- Drop the image on the page, choose it with the file picker, or paste it from the clipboard.
- The check runs inside your browser; the image is not uploaded to us.
- Read the headline first. Fixed rules choose it, not a score.
- Read the sentence under it. It names the certificate and any trust list that vouches for it.
- Read the findings, each labelled by its weight, and the "Not checked here" line every result shows.
What does each result headline mean?
This table covers the headlines that concern a signed record, plus the one you see most often without one.
| Headline | What it means |
|---|---|
| Trusted Content Credentials found | The record validates and its signer is on the current C2PA Trust List. The record does not say whether AI was used. |
| Trusted Content Credentials declare AI use | Same trust, and the signed record says a generative source was used. This is the record's claim, not ours. |
| Trusted Content Credentials declare another source | Same trust, and the signed record names a source that is not generative AI. |
| Content Credentials found; legacy trust-list match | The record validates, but the signer is on the frozen Interim Trust List, not the current list. |
| Content Credentials found; signer trust not established | A record is present and matches the image, but neither list vouches for the certificate, so its name is not verified. |
| Content Credentials failed validation | The record does not validate, or the image changed after it was signed. Do not rely on the record. |
| Content Credentials are not available locally | The file only points to a record stored online. We never fetch it. |
| Content Credentials check incomplete | A record is there, but the check could not finish, so trust or the signature is unconfirmed. |
| No supported provenance signals found | Every check ran and found nothing it reads. This does not mean the image is real or human-made. |
What is the difference between "trusted" and "signer trust not established"?
A signature proves that someone with a certain certificate signed the record and that the bytes match. Whether the certificate belongs to who it says is a separate question, and only a trust list answers it. The Content Authenticity Initiative's documentation says the Interim Trust List was frozen as of January 1, 2026, and that validators may consult both lists but must distinguish between credentials signed under each. That is why we show two different headlines.
Our copy of the current list is dated October 17, 2026, and the Interim list's anchors file last changed January 8, 2026. A signer added later is not recognised until we refetch. See why a signer is not trusted and the two trust lists.
What did we find on real files?
We ran 32 real image files through the checker: 29 from Wikimedia Commons categories whose names say they hold images generated by a named tool, 2 samples published by vendors and 1 C2PA test image. The category names are the sources' claims, and we cannot know how each file was made or handled before upload. Of the 32, 9 carry a Content Credentials record. One, a ChatGPT file, reads "Trusted Content Credentials declare AI use". Five read "Content Credentials found; legacy trust-list match" (Adobe Firefly and Microsoft Image Creator files). Three read "Content Credentials found; signer trust not established" (two ChatGPT files and the C2PA test image).
For the 9 real files with a record, the check took a median of 42 milliseconds and at most 176, the slowest being the first file, which also loaded the engine and both lists. During the 50-file run (the 32 real files and 18 we made) the test page made 0 requests to any other site. The counts and per-file results are on our accuracy page; the timing and network figures are not shown there. We measured on October 30, 2026, in one browser, on a small set of files. They show what the checker read in those files, not what any product does. A "Trusted" headline appears for one real file we hold; for the files we made, only with a test trust list loaded.
Can I check with another tool as well?
Yes, when the answer matters. Verify from the Content Authenticity Initiative inspects a file you drag in. It also reads video, audio and PDF files, and it warns that Content Credentials are still rolling out, so a file may have nothing to show. Adobe's Inspect accepts the same formats and adds C2PA, GIF and JXL.
Both are other companies' sites; read their terms for how they treat your file. When a file only points to an online record, our checker does not fetch it, and our result says Inspect can look it up.
To practise on known files, the C2PA project publishes public test files, with "bad" files included where possible.
What this means when you check an image
Open the checker; read the headline.
- "Content Credentials found; legacy trust-list match": a record validates and the signer is on the frozen Interim Trust List. It tells you who signed, by that list, not that the record is true.
- "Content Credentials found; signer trust not established": a record matches the image, but no list vouches for the certificate. Files from real companies land here, and a forged record can read the same.
- "Trusted Content Credentials found": the signer is on the current C2PA Trust List. The record does not say whether AI was used.
- "No supported provenance signals found": the checker read nothing it supports. Most images lose these signals when shared or re-saved, so it says nothing about how the image was made.
Every result carries the line "These checks cannot determine whether an image is AI-generated." A valid record shows what the signer declared. Weigh that, and the source of the file, yourself.
Guides on this topic
- What are Content Credentials?
- How to read a Content Credentials record
- What "signer not trusted" means
- The C2PA Trust List and the Interim Trust List
- Can Content Credentials be faked?
Frequently asked questions
- Is my image uploaded when I use the checker?
- No. The check runs in your browser. In our 50-file test run, the test page made 0 requests to any other site. Following a link to another company's checker is separate: you upload the file there.
- Why does a real company's file say "signer trust not established"?
- Because neither trust list we load vouches for that certificate. Two older ChatGPT files in our set read this way: their certificates had expired. The headline cannot tell such a file from a record signed with a self-made certificate.
- Does a trusted result prove the image is real?
- No. It shows who signed and that the image matches the record. The record's claims are the signer's declarations.
- What if the checker finds nothing?
- The record may never have existed, or it may have been lost in a screenshot, a re-save or an upload. Finding nothing does not mean the image is real or human-made.