What are Content Credentials (C2PA) and how do you check them?

What are Content Credentials, and what is C2PA?

They are a provenance record stored inside the file itself. The problem they address is simple: you cannot tell where an image came from by looking at it. A real photo and a generated picture can be equally convincing, and a genuine photo can be retouched without leaving a visible trace.

C2PA is the open technical standard that defines how the record is written. Content Credentials is the public-facing name for it. In everyday use the two terms point to the same thing: a manifest attached to the file.

That manifest can include:

• Which device or software created the file.
• Whether it was fully or partly generated with AI, when the tool declares it.
• Which edits were made, such as cropping, color adjustments or combining it with other files.
• Who signed the record: the camera maker, the editing app or the organization that published it.

What separates it from ordinary metadata is the digital signature. The manifest is bound to the file's content and signed with a certificate, so a verifier can tell whether the file or the record changed after signing.

How do Content Credentials get attached to a file?

The tool that creates or modifies the file adds them, provided it supports the standard and the feature is switched on. There are three usual entry points:

• Capture: some cameras and phones sign a photo at the moment it is taken.
• Editing: some apps add a record of the changes on export and keep the earlier manifest as history.
• Generation: some image, audio and video generators mark their output as AI-made.

Support is partial. Most devices and apps in use today write no credentials at all, and in several that can, the feature is optional. The chain also breaks easily: a screenshot, a format conversion, or a trip through a social network or messaging app that does not preserve the data leaves the file with no record.

How do you check whether a file has Content Credentials?

You need the original file, or the most direct copy you can get. A screenshot has nothing left to check.

1. Get the file as its author delivered it or as it downloads from the source, without forwarding it through a messaging app.
2. Load it into a C2PA-compatible verifier, such as the one on the Content Credentials site.
3. Read who signed the record. A signature from a recognizable maker or organization carries more weight than one from an unknown issuer.
4. Go through the history: what created the file, whether AI use is declared, and which edits are listed.
5. Compare the record with what is being claimed about the image: date, author and context.

Some platforms show a label or icon when they detect this data. That helps, but open the details anyway: the label summarizes, the manifest explains.

What does each result mean?

There are three possible outcomes and each one reads differently. The most common mistake is treating the second as a bad sign.

Invalid credentials call for caution, not an accusation. They mean the record no longer vouches for the file in front of you, so it stops being useful as a signal of origin.

What do Content Credentials not tell you?

They do not tell you whether what the image shows actually happened. Valid credentials back the provenance the signer declares: what made the file and which steps it went through. A camera-signed photo can show a staged scene, and a genuine image can be posted with a false date or location.

They do not necessarily identify a person either. Often the signer is the device or the app, not the photographer. And trust in the record depends on trust in the signer: the standard guarantees the record was not altered, not that whoever issued it told the truth.

How do provenance, detection, metadata and a timestamp fit together?

Each technique answers a different question, which is why they work better in combination. None of them settles on its own whether an image can be trusted.

To go deeper into the other pieces, see how to tell if an image is AI-generated and what a trusted timestamp is. To read what a file declares about itself, use the metadata viewer.

What does our authenticity report do with a file?

The authenticity report gathers several of those signals into one document. It is worth knowing exactly what it includes and what it does not:

• An automatic origin estimate. It indicates whether the file looks made by a person or with AI, with a confidence level and its limits. That is guidance, not proof.
• Presence of a C2PA manifest. The report detects whether the file carries one. It does not validate the signature; for that you need a credentials verifier.
• A metadata reading. It lists what the file declares, labeled as declared data.
• Fingerprint and timestamp. It computes the file's digital fingerprint and seals it with a date, so you can later show that this same file existed at that moment.

The report documents which file was reviewed, when, and with what result. It does not certify that the image is true, and it does not replace a forensic examiner.

What should you check before trusting where an image came from?

1. Get the original file, not a screenshot or a forwarded copy.
2. Load it into a credentials verifier and note the result.
3. If the credentials are valid, read who signed them and which edits are declared.
4. If there are none, draw no conclusion and move on to the other signals.
5. Check the metadata and look for the earliest post of the image.
6. Use an automatic detector as one more clue.
7. If you need a record, keep the file unmodified and timestamp its fingerprint.

Common mistakes

• Reading missing credentials as a sign of fakery. It is the normal case, not an anomaly.
• Confusing provenance with truth. The record describes the file, not the events.
• Checking a screenshot. The manifest does not survive one.
• Stopping at the platform label. Without the details you do not know who signed or what was edited.
• Assuming that detecting a manifest means validating it. Its existence does not guarantee a correct signature.