Content Credentials Checker

Content Credentials versus ordinary metadata

Ordinary EXIF and XMP fields can name a camera, a lens, an editor, or a creator, and you can read all of them in the EXIF viewer. Anyone with a metadata editor can rewrite them in seconds, and most sharing platforms delete them without asking. They are useful context and worthless as proof.

Content Credentials package the same kind of information into a signed C2PA claim that is cryptographically bound to the asset bytes. A validator can therefore detect both substitution of the record and modification of the asset. That is the entire difference, and it is a large one.

The practical consequence: when a credential and a metadata field disagree, the credential is the stronger evidence. When only metadata exists, you have a hint that deserves corroboration.

How the ingredient chain preserves history

The design goal that makes Content Credentials interesting is that editing does not have to erase provenance. When a C2PA-aware editor opens a file that already has a credential, it can record the original as an ingredient of the new manifest rather than discarding it.

Over several rounds of editing this builds a chain: an original capture, a raw conversion, a retouch, a crop for publication. Each link is separately signed, and each declares what it consumed and what it did. A validator can walk backwards through that chain.

This is why an image that has been heavily edited can still carry meaningful provenance, and why the ingredient list is often more informative than the top-level verdict. An asset whose earliest ingredient declares a generative source type was AI-originated even if every later action was ordinary photo editing.

Reading digital source types without overclaiming

Digital source type is a controlled vocabulary term that describes where the content came from. The values that matter most in practice distinguish content captured by a camera, content produced entirely by a generative model, content where a model modified an existing asset, and content composited from several sources.

The distinction between fully generated and model-modified is the one most often lost in reporting. An image where a generative tool removed a background or extended a canvas is not the same thing as an image invented from a prompt, but a careless reader will treat both as 'AI-generated'.

When the source type is present, quote it precisely. When it is absent, say it is absent rather than inferring one.

• Captured — recorded by a device from the physical world.
• Trained-algorithmic media — produced by a generative model.
• Composite with trained-algorithmic media — a mix of captured and generated material.
• Algorithmically enhanced — an existing asset processed by a model.

Trust lists and why a valid credential can still look unverified

Signature verification and trust are separate questions. A signature can verify perfectly while the certificate that produced it is unknown to the validator, because trust depends on which certificate authorities the validator has been configured to recognise.

A browser-based validator ships with a limited trust configuration. Organisations running their own newsroom pipeline typically configure their own list. So a credential that shows as valid-but-untrusted here may be entirely legitimate and recognised elsewhere.

Read the signer identity in the evidence panel and judge it yourself. A signature from an organisation you recognise is meaningful even if this particular validator has never heard of them.

What to do when there are no credentials at all

Most images on the open web carry nothing. That is the baseline, not an anomaly, and it is why this tool refuses to convert emptiness into a verdict.

The productive next steps are outside provenance: reverse image search to find earlier copies, checking whether the ordinary metadata survived, looking at whether the account or publication has a track record, and asking the source directly for an original file rather than a re-shared copy.

An original file straight from the creator's device or export is far more likely to retain both credentials and metadata than anything that has passed through a social platform.