How to show a photo is real when someone says it is AI
No tool can prove a photo is real. You can keep evidence that fits your account: the original files with their dates, a signed capture record if your camera makes one, and your edit history. Each shows something specific, and none settles the question alone. A detector's flag is one signal, not a verdict.
Why would a detector flag a photo its owner says is real?
Full Fact reported on October 21, 2026 that Google's public SynthID Detector described five images shared by politicians as made or edited with Google AI. The people who posted them said no AI was used. Full Fact calls these "likely" false positives, because it cannot rule out a small edit such as brightening, and it says it did not measure how often this happens.
Google told Full Fact that using the detector without prior suspicion of AI editing raises the chance of an occasional false positive, and that results should be read alongside other signals. After a Google update on October 15, Full Fact says the tool labelled the five images "Unsure", though it reports another image was still flagged. Full Fact's advice is to "always check for other signs too".
So a flag is a reason to look at the evidence, not the evidence. If someone accuses you, ask which tool they used and what its own page says about errors. OpenAI's guide to its checker says a result of not detected is an absence of detected evidence, not proof the content is human-created.
What evidence can a photographer keep?
| Evidence | What it shows | What it does not show |
|---|---|---|
| The original file, such as the RAW file or the camera's own JPEG | The file as it left the camera, which a reviewer can compare with the version you posted. | That nobody else holds a similar file. Keep it unedited and work on copies. |
| File dates and camera metadata (EXIF) | The dates and camera model the file says. With other frames from the same shoot, they give a reviewer a sequence to check. | That the fields are true. Ordinary software can edit them; see can EXIF data be faked. |
| A signed capture record, where your camera offers one | Who signed, and that the image still matches what was signed. | That the scene was as the record says. A name is vouched for only if a trust list covers it. |
| Edit history, such as project files and a record written by editing software | The steps taken between the capture and the posted version. | That the history is complete, if some step happened in another tool. |
Do not re-save, screenshot or convert the originals. Store them untouched.
Which cameras and phones support Content Credentials?
- Google Pixel 10. Google's September 10, 2025 post says the Pixel 10 lineup has Content Credentials built in across every photo created by Pixel Camera. The same post says any claim's security depends on the integrity of the app and the operating system.
- Leica M11-P. The Content Authenticity Initiative's how-it-works page describes it as the world's first camera with Content Credentials built in.
- Samsung Galaxy S25 series. Samsung's February 7, 2025 post says the lineup supports Content Credentials, adopted for content created and edited with generative AI. It does not say ordinary camera photos are signed.
This is not a full list; our guide to cameras and phones with Content Credentials covers more makers. The Initiative adds that edits made with software that supports Content Credentials can be recorded too.
What does a signed record show, and what does it leave out?
The C2PA explainer says Content Credentials do not give value judgments on whether the provenance data is true. The Initiative says the same: Content Credentials "aren't intended to prescriptively indicate whether a piece of content is 'real'".
A name inside a signature shows less than it seems to. Our measured result: we signed one existing image with a certificate we made ourselves, naming "Leica Camera AG" and claiming a Leica M11 capture. Our checker reads it as "Content Credentials found; signer trust not established", because only a trust list vouches for a signer. We ran this on October 30, 2026; see our guide on whether Content Credentials can be faked.
Camera makers do appear on the lists we load. On October 30, 2026 we read the names in all 174 certificates in three list files. Leica, Sony, Nikon, Canon, Fujifilm and Samsung each appear as an anchor on the Interim Trust List, frozen since January 1, 2026 (copies fetched October 27, 2026). A list holds signers, so a name on it does not show that any model signs by default. We hold no camera-signed file, so we cannot say what a given camera's photo will read.
Will the record survive when I share the photo?
Not in our test. On October 30, 2026 we took the 9 real files in our set that carry a Content Credentials record and redrew each one on a browser canvas, which keeps pixels only. Saved as PNG, 0 of 9 kept a record. Saved as JPEG, 0 of 9 did. All 18 outputs read "No supported provenance signals found". A canvas is not a platform, and we contacted none. See our accuracy page and the guide on screenshots and metadata. Keep the original: a record can be gone from the copy people argue about.
What this means when you check an image
Open the checker, choose your original file, and read the headline. Three matter here.
- "Trusted Content Credentials found". A signed record is present, its signer is on the current C2PA Trust List, and the record does not say whether AI was used. It shows who signed, not that the scene was as described.
- "Content Credentials found; signer trust not established". A record is present, but no list we load vouches for the signer. The detail line says whether the image matches the record. The name shown is only what the certificate says.
- "No supported provenance signals found". The checker found none of the signals it reads. Saving, screenshotting or sharing often removes these signals, and the result says nothing about how the photo was made.
Under every result the checker says: "These checks cannot determine whether an image is AI-generated." It gives no probability. Its result page also links to Google's SynthID Detector, other companies' checkers and a reverse image search. Those open other sites, you may need to sign in, and your image goes to them.
Guides on this topic
- What are Content Credentials? is the main page on this topic.
- Cameras and phones with Content Credentials covers capture.
- How to add Content Credentials to your own images shows what a self-signed record reads.
- Can Content Credentials be faked? covers forged certificates.
- Why AI image detectors disagree explains what a percentage means.
Frequently asked questions
- Can I prove my photo is not AI-generated?
- No tool can prove it. You can show evidence that fits your account: the original file and its dates, other frames from the same shoot, a signed capture record if your camera made one, and your edit history.
- Is a detector flag evidence against me?
- It is one signal. Full Fact reported five apparently real images that Google's detector flagged, and Google said results should be read alongside other signals.
- Should I add Content Credentials to my own photos?
- You can add a record, but a certificate you make yourself is on no trust list, so our checker reads it as "Content Credentials found; signer trust not established".
- Does a photo with no record look suspicious?
- Not on that fact alone. The C2PA explainer says adding provenance is optional, and that its specifications do not intend to make media without Content Credentials universally less trusted. A camera may not sign, and sharing can remove a record, so a missing record does not say how the photo was made.