Content Credentials for Readers: What the Label Can and Cannot Tell You

You may have started to see a small label on some photographs and news images, sometimes a pin-shaped icon in a corner, that opens to show where the picture came from. That label is a Content Credential, and it is the visible end of an open standard called C2PA. This guide is written for readers rather than for the companies that make the tools: what the content credentials label can tell you about an image or an article, how it is protected, how to look at one, what removes it, and why an image without one is not automatically suspect.

What a Content Credentials label tells you

A Content Credential is the plain name for what the standard calls a C2PA manifest: a set of signed statements, called assertions, about a file. The C2PA explainer lists the kinds of things a manifest can assert: where and when the content was created, what modifications were made and with which tools, and how AI was used in making it. Publishers can add their own assertions, such as who created the content or whether they permit it to be used for AI training.

Read one and you might learn that a photograph was captured on a particular camera model, opened in an editing application, cropped and colour-corrected, and published by a named news organisation on a given date. Or you might learn that an image was generated by an AI tool from a text prompt. What you will not learn is whether the caption is accurate, whether the scene was staged before the shutter opened, or whether the news organisation is one you should trust. The explainer is direct about this: the technology makes no judgement about whether the provenance is true, only whether the record is well formed, untampered, and signed by someone on a known trust list.

How the C2PA standard signs a manifest

The mechanism is worth knowing in outline, because it explains both the strength of the label and its fragility. When a credential-aware camera or application creates or edits a file, it writes a manifest containing the assertions and a cryptographic hash of the content, a short fingerprint that changes if any pixel changes. The manifest is then signed with the private key of the device or software, and the signature can be checked with the matching public key.

The C2PA explainer describes the hashing as a chain: hashes of each part are folded into further hashes and the whole is signed using standard X.509 certificates, the same family of certificates that secures websites. Change the image, the assertions or the signature and the pieces no longer match, so the record is tamper-evident. The manifest is usually embedded inside the file. The standard also allows a soft binding, an invisible watermark or a fingerprint lookup, so that a manifest can sometimes be found again after it has been stripped.

Each edit in a credential-aware tool adds a new manifest that refers to the previous one, so a fully documented file carries its history as a chain. The coalition behind the C2PA standard, whose members include camera makers, software companies such as Adobe and Microsoft, and news organisations, publishes the specification openly and royalty free; anyone can build a tool that writes or reads these records.

How to inspect a content credential

1. Look for the label on the page. On sites that support the standard, an icon sits on or beside the image, and clicking or hovering opens a panel that summarises the credential. BBC News began publishing images with these labels in March 2024, and its research department has described how the panel looks on its site.
2. Open the full credential. The panel usually links to a fuller view. Read the signer's name, the date, the list of tools and edits, and any statement about AI. Note which assertions are present and which are absent.
3. Check the file with a verify tool. If you have the file itself, upload it to a verification page run by the coalition or by a publisher, or use one of the open-source command line tools built on the standard. The tool reports whether the signature is valid, who signed it, and the chain of edits.
4. Read the file's other metadata. Standard photo metadata defined by the IPTC (creator, copyright notice, credit, caption) lives in the same file and is not signed, but it is often filled in and worth reading alongside the credential. The IPTC's own metadata viewer shows it.
5. Decide with context. Combine what the credential says with what you know about the signer and with ordinary checks: where the image was first published, whether it appears elsewhere with a different story, and whether the caption matches what is in the frame.

What the credential can show and cannot show

The point about missing edits matters. The explainer notes that if a file is cropped in a tool that does not understand credentials and then brought back into one that does, the new signer implicitly vouches for the earlier crop. The record is then still trustworthy to the degree the last signer is, but it is not complete, and a careful reader keeps that distinction.

What strips or breaks a credential

A screenshot has no credential, because a screenshot is a new image of your screen, not a copy of the file. Re-saving an image in software that does not understand the standard usually discards the manifest, and re-saving it in software that does understand the standard but after an unrecorded edit produces a mismatch, which a verify tool reports as invalid or broken. Format conversion can do the same.

Social platforms are the largest gap. The IPTC's 2019 tests of photo metadata across about a dozen social sites found that most removed at least some embedded metadata on upload or download, and that rescaling an image was a common point of loss. Credentials are newer than the metadata tested then, and support has been announced by some platforms since, but the safe assumption for a reader is that an image copied through a social feed has lost its history unless the platform says otherwise. This is why soft bindings exist, and why the original publisher's page is the place to look.

What the absence of a credential does and does not mean

Most images on the web today have no credential, and the standard's own FAQ says that no assumption about trustworthiness should be made purely on that basis. Adding provenance is optional; the explainer says the intention is not a two-tier web where unlabelled media is treated as suspect. An unlabelled image from a publisher you already trust is exactly as trustworthy as it was before the standard existed.

What absence does mean is that you have to do the work the other way: find where the image was first published, compare versions, and read what the publisher says about it. Treat a label as a shortcut through some of that work, not as the work itself. And a credential from an unknown signer tells you little more than no credential at all.

The reader's side of provenance

There is a parallel here with the rest of this site. A Content Credential is the publisher's record of how a file came to be. A reader's record is the same idea from the other side: a highlight on the passage you relied on, a dated note saying what you took from it, and a link that opens the page at that line. A text fragment link made with the highlight tool is the reader's signature on a passage: it fixes which words, on which page, at which date, you are vouching for.

The two records meet in practice. When you cite an image, note whether it carried a credential and what it said. When you quote a claim, run it through the source verification workflow for text and log the result. And when your own work used AI in some part, say so plainly, in the same spirit as the AI assertion inside a manifest; stating which parts of a piece had AI help is a credential you write yourself. For the claims inside an article rather than its images, checking an article's statements against the sources it cites is the reader's equivalent of opening the manifest.

Pitfalls when reading Content Credentials

• Treating the label as a truth mark. It marks integrity and origin. A signed image of a staged scene is still a staged scene.
• Reading a valid signature as a trusted signer. Validity means the record is intact. Trust comes from knowing who the signer is.
• Assuming a missing edit did not happen. Tools that do not write credentials leave no trace. The record shows what was recorded, not everything that occurred.
• Judging an image by its screenshot. The screenshot never had a credential. Find the original file or page.
• Counting absence against an image. Most of the web is unlabelled. Absence sends you to ordinary checks; it is not a verdict.

A credential describes a file's history; your own record describes your reading of it. The two meet in the section on highlights as evidence: logging a verified source with a highlight link is what you do once the credential has told you where the file came from, and citing the exact line you rely on is how the check ends up in your work.

Content credentials give readers something they have never had for images: a signed, checkable history of the file. Read the label for what it is, an account of origin and edits from a named signer, and keep your own record on your side of the screen, a highlight, a date and a link, so that what you took from the image is as checkable as the image itself.