Guide
What Content Credentials Actually Tell You
Last updated October 6, 2026
"Does this image have Content Credentials?" is a reasonable question. "Is this image real?" is a different one, and Content Credentials do not answer it. This guide explains what the technology records, what its checks mean, and where people commonly read too much into them. It is based on the C2PA technical specification, version 2.2 (May 2025).
What is inside
C2PA stands for the Coalition for Content Provenance and Authenticity. A file can carry a manifest: a bundle of assertions (statements, such as "captured by this camera", "edited with this software", "this was generated by an AI model"), a claim that references them, and a signature over the claim made with the signer\'s certificate. A hard binding, a cryptographic hash of the image data, ties the manifest to that exact file. When a file is edited, a new manifest can be added that points to the earlier one as an ingredient, so the file carries a chain of steps.
What a check can confirm
- Whether credentials exist in the file at all.
- Whether the signature on the claim is mathematically valid.
- Whether the hash in the credentials still matches the image data. If not, the image changed after signing.
- What the signer chose to declare: tool, edit actions, a source type such as a camera capture or an AI model.
The C2PA Content Credentials Checker shows these things in plain language and keeps findings about "the data was altered" apart from findings about "the signer is not on a trust list".
What a check cannot confirm
- That the content is true. The specification says validators should not make value judgements about whether provenance data is "good" or "bad"; they only report whether the assertions can be validated. A signed statement can be accurate about a misleading picture.
- Who the signer really is. Trust in a signer comes from trust lists and certificate rules. A tool that does not apply a trust list shows signers as stated.
- That an image without credentials is fake. Most images have none. Cameras and apps rarely add them, and screenshots, re-saves, resizing and many uploads remove them.
- That an image with a missing "AI" declaration is not AI. An AI tool may declare its output, but nothing forces a person to keep that declaration, and nothing here detects generated content from its pixels.
Reading common results
| You see | It means |
|---|---|
| No credentials | Nothing was found. It says nothing about the image\'s origin. |
| Credentials, valid signature, hashes match | The record belongs to this file as it is now. Who signed it and whether to trust them is a separate question. |
| Credentials, hash mismatch | The image was changed (or re-encoded) after signing. The credentials describe an earlier state. |
| Credentials, signer untrusted | The certificate is not on a trust list that was used. Test and self-signed certificates look like this. |
Practical advice
Treat credentials as one clue next to the source, the context and other evidence. Inspecting a file locally avoids uploading it: the checker above reads the file in your browser and has network features of the reader switched off. To see other hidden data in a photo read what image metadata reveals and use the Image Metadata Remover; note that removing metadata also removes embedded credentials. For file-level clues such as compression history, the JPEG Quality Estimator shows how a JPEG was encoded, which is a hint, not proof.