How to Verify C2PA Content Credentials (4 Methods, 2026)

Four ways to verify C2PA Content Credentials on an image, what the manifest shows, what a failed signature means, and which phones and cameras sign at capture.

C2PA Content Credentials are tamper-evident metadata attached to an image that record how it was made and edited, including whether AI was involved. We go deeper on this in whether removing C2PA makes an AI image undetectable. To verify them, upload the file to a Content Credentials viewer like the official one at contentcredentials.org, which reads the signed data and shows the image's history. When credentials are present they are strong proof of origin. When they are missing, which is common, you fall back to an AI image detector and the visual tells. Here is how the whole thing works.

Key Takeaways

• C2PA Content Credentials are signed, tamper-evident metadata that record an image's origin and edit history.
• You verify them by uploading the file to a Content Credentials viewer, which shows what the signature contains.
• Major tools attach them: OpenAI adds credentials to DALL-E images, and camera and editing apps are adopting the standard.
• Credentials are fragile. Screenshots, crops, and most social uploads strip them, so absence is not proof of anything.
• When credentials are gone, use pixel-based detection plus provenance signals like Google SynthID together.

What C2PA actually is

C2PA stands for the Coalition for Content Provenance and Authenticity, an industry group that built an open technical standard for recording where a piece of media came from. We wrote about EU AI Act Article 50 image labelling rules in more depth. The C2PA standard defines how to attach a signed manifest to a file: who created it, what tool made it, whether it was AI-generated, and what edits happened along the way.

Content Credentials is the consumer-facing name for this data. Think of it as a nutrition label for an image. When it is intact, you do not have to guess whether a picture is real, because the file carries a signed record of its own history.

Why signed and tamper-evident matter

Ordinary metadata is easy to fake. Anyone can edit an EXIF field. C2PA is different because the manifest is cryptographically signed. If someone alters the image or the record after signing, the signature breaks, and a verifier will show the credential as invalid rather than trusting it.

That is the whole point. A signature you cannot forge turns metadata from a weak hint into real evidence. It does not stop someone from stripping the data entirely, but it does stop them from quietly rewriting it and passing it off as genuine.

How to verify Content Credentials

1. Get the original file at full quality. A screenshot will almost never carry the credentials.
2. Open a Content Credentials viewer. The official verifier at contentcredentials.org is the simplest.
3. Upload or drag the image in.
4. Read the manifest. It will show the issuing tool, whether AI was used, and the edit steps recorded.
5. Check the signature status. A valid signature means the record is trustworthy; a broken or missing one means treat it with caution.

Some platforms also show a small credentials icon on images that carry the data, which you can click to see the same history inline.

Four ways to verify, from easiest to most thorough

The official Verify site is one route. There are three more, and they suit different situations.

1. The Verify site (contentcredentials.org). Drag the file in and read a plain-language summary. It can also consult Adobe's public Content Credentials cloud, which sometimes recovers a record for a file whose own metadata was stripped. The file is uploaded to do this.

2. An in-browser C2PA viewer. Sites such as C2PA Viewer run the check with WebAssembly in your browser, so the file never leaves your machine and you get the raw manifest JSON. Better for anything confidential.

3. A browser extension. The Content Credentials extensions for Chrome, Edge and Brave add a right-click "Verify Content Credentials" option on any image, video or audio file on a page. Fastest for checking something you found while browsing, with the caveat that most page images have already been re-encoded by the site.

4. c2patool on the command line. The open-source tool from the Content Authenticity Initiative prints the full manifest, the certificate chain and the validation status. Run c2patool photo.jpg for the summary, or add --detailed for everything. This is the route for batch checks and for anyone who needs the validation codes rather than a green tick.

What the manifest actually shows

A manifest is a small signed record inside the file, and every verifier is reading the same fields:

• Claim. Who made the file, which software or device, and when.
• Assertions. What happened: created, edited, cropped, colour-corrected, published. The field that answers the AI question is digitalSourceType. A value of trainedAlgorithmicMedia means fully AI-generated; compositeWithTrainedAlgorithmicMedia means a real image with AI-generated parts.
• Ingredients. Earlier files this one was built from, each with its own manifest, which is how an edit chain is recorded.
• Signature. A certificate from a recognised issuer, checked against a trust list. Adobe, OpenAI, Google and the camera makers sign with their own certificates.
• Hard binding. A SHA-256 hash of the content. Change one pixel after signing and the hash no longer matches.

The C2PA specification, now at version 2.2, defines the format. The Content Authenticity Initiative publishes the open-source tools.

Reading a failed check

Three different results get lumped together as "it failed", and they mean different things.

No manifest found. The file has no credentials. This is the normal state for most images online, because screenshots, format conversions and most upload pipelines strip them. It says nothing about whether the image is AI.

Manifest present, signature invalid. The file was changed after it was signed, or the signing certificate is not on the trust list. Treat the record as untrustworthy, and treat the image as edited by someone who did not re-sign it.

Manifest present, signature valid, ingredients missing. The current step is verified but the history before it is not. Common when a signed image was edited in a tool that supports C2PA but the original ingredient was not carried over.

Only the second result is evidence of tampering. The first is silence, and silence is where the AI image detector takes over.

Which phones and cameras sign at capture

Provenance is only useful if real photos carry it too, and that started in earnest in 2025. The Google Pixel 10, announced in September 2025, signs photos at capture. The Samsung Galaxy S25, announced in January 2025, does the same. Leica's M11-P and Sony's Alpha line, including the a9 III and a1 II, sign in camera, and Nikon has shipped support on selected bodies. Adobe Photoshop and Lightroom write credentials on export when the option is enabled.

That list will be out of date within months, which is the point: the more capture devices sign, the more often a missing credential on a supposedly straight-from-camera photo becomes a question worth asking.

When credentials are present

If a picture arrives with intact, valid Content Credentials, you are in the best case. You can see, for example, that an image was generated by DALL-E, or that a real photo was edited in a specific app. This is far stronger than any detector score, because it is signed data rather than a statistical guess.

This is why we treat provenance as the top of the evidence stack. For a generator like DALL-E, credentials plus the visual style make identification easy, which we cover in the DALL-E image detector guide.

When credentials are missing

Here is the catch, and it is a big one. Content Credentials are fragile. A screenshot drops them. A crop can drop them. Most social platforms strip metadata on upload to save space and protect privacy. So the majority of images you meet in the wild will have no credentials at all.

Absence is not evidence. A missing credential does not mean an image is real, it usually just means the file was processed somewhere along the way. When that happens, you switch to pixel-based detection. Upload the image to GPTOne's AI image detector for a confidence score and a heatmap of the regions that look generated. That reads the image itself rather than its metadata, so it still works after the credentials are gone. We explain the strengths and limits of that approach in do AI image detectors actually work.

C2PA versus SynthID

People mix these up, so here is the clean distinction. C2PA is metadata attached to the file. Google SynthID is an invisible watermark embedded in the pixels of some AI images, mainly from Google's own models. SynthID survives some edits that would strip C2PA, because it lives in the image data rather than the file wrapper.

They complement each other. C2PA tells you the file's recorded history; SynthID can flag Google-generated content even after metadata is gone. A thorough check looks for both, then adds detection on top. We go deeper on the watermark side in our SynthID guide.

How we test provenance and detection together

We do not just theorize about this. We ran a structured 60-image benchmark across five categories, mixing AI and real photos, and documented the method in our AI image detector benchmark. The practical lesson: provenance and detection are layers, not rivals. Use signed credentials when they exist, use detection when they do not, and weigh the false-positive risk in both.

That layered habit is what separates a confident verdict from a guess. Check the credentials, run the scan, read the visual tells, and let the three agree before you trust an image that matters.

Who is adopting C2PA

The standard only works if the tools people use actually support it, and adoption is growing fast. OpenAI attaches Content Credentials to DALL-E images. Adobe builds them into its creative apps. Camera makers including Leica and Nikon have shipped or announced support so photos can be signed at capture. Microsoft, Google, and major news organizations back the coalition too.

That momentum matters, because provenance is a network effect. The more cameras that sign at capture and the more platforms that preserve credentials instead of stripping them, the more often you will actually have this evidence when you need it. We are not there yet, but the direction is clear, and it is why learning to read Content Credentials now is worth the few minutes it takes.

Where verifying credentials really helps

Some situations make provenance genuinely decisive. Journalism and fact-checking, where a signed capture record can confirm a photo came straight from a camera. Legal and insurance work, where an edit history can show whether an image was altered, a case we walk through in the insurance claims guide. Buying and selling, where a credential on a product photo adds trust. And any high-stakes share during breaking news, covered in how to verify a viral news image.

In each case the pattern is the same: if credentials survive, they settle the question fast. If they do not, you drop down to detection and the visual tells. Knowing which situation you are in is half the skill, and it saves you from over-trusting a blank result or under-trusting a valid one.

FAQ

What are C2PA Content Credentials in simple terms?

They are a signed, tamper-evident record attached to an image that shows how it was made and edited, including whether AI was used. Think of it as a verifiable nutrition label for a photo.

How do I check an image's Content Credentials?

Upload the original file to a Content Credentials viewer such as the official one at contentcredentials.org. It reads the signed manifest and shows the tool, AI usage, and edit history.

Why do most images have no Content Credentials?

Screenshots, crops, and most social media uploads strip metadata. So a missing credential is normal and does not prove an image is real. Fall back to an AI image detector in that case.

Is C2PA the same as a watermark?

No. C2PA is metadata attached to the file. Google SynthID is an invisible watermark inside the pixels. They complement each other, and a full check looks for both.

What should I do if the credentials are missing?

Run the image through a free AI image detector for a confidence score and heatmap, then check the visual tells. Combine those signals rather than trusting any one alone.

When the credential is missing, the GPTOne AI image detector reads the pixels and the C2PA data in one upload, with a region heatmap on the free tier.