How can you tell if a photo was made with AI?
There is no single test that gives a definitive answer. But many AI tools leave traces in the files they create: signed Content Credentials, an invisible watermark or the tool's name in the file metadata. If one of these traces is found, it is a strong sign. If none is found, that does not prove the photo is real.
1. Find the original file
Most traces are stored inside the file, and some of them are lost every time the file is shared. Apps such as Instagram, X, Facebook and WhatsApp recompress uploaded images and often strip their file metadata. Taking a screenshot removes everything stored inside the file.
So, if you can, download the original file from the place where the photo was first posted. A file downloaded from social media will very likely show no traces, and that proves nothing.
2. Look for Content Credentials (C2PA)
Content Credentials (technically called C2PA) are a signed record added to a file. Tools from OpenAI, Adobe, Google and Microsoft add this record to the images they generate; some cameras and phones also add it to the photos they take.
The record can include which tool created the content and a statement that it was “generated with AI.” If the signature is valid and the signer is a recognized organization, this is the strongest evidence. If the file was changed after it was signed, the signature no longer holds, and that is shown separately.
3. Check for invisible watermarks
Some companies embed a watermark in an image's pixels that the eye cannot see. The most common one is Google's SynthID watermark. It is designed to stay in place in most cases, even if the file metadata is removed and the image is compressed.
You can check for SynthID only with Google's own tool. Open-source watermarks, such as Adobe's TrustMark, can be read by anyone.
4. Look at the file metadata, but don't trust it too much
Photos usually carry information called EXIF and XMP: the camera model, the date taken and the software used. Some AI tools write their own name or the image's generation settings here.
This information is unsigned; anyone can add or remove it in a few clicks. A file with camera information may still have been made with AI. That is why file metadata is only an indicator, not evidence.
5. Looking closely is no longer enough
A few years ago, mangled hands, garbled text and inconsistent shadows gave AI images away. Current models make these mistakes far less often. Something that looks odd can be a reason for suspicion, but on its own it is not enough to decide.
What do the results mean?
Our checking tool gives each file one of these results:
- AI trace found: There is an intact AI statement signed by a recognized organization.
- AI indicators found: There is information pointing to AI, but it is unsigned or cannot be verified.
- Content Credentials broken: The file was changed after it was signed.
- Camera information found: There is camera information and no AI trace. This does not prove the photo is real.
- A trace was found but its source could not be verified: There is a trace, but where it came from could not be determined.
- No traces found: There is no readable trace. This does not prove the photo is real either.
Frequently asked questions
If no traces were found, is the photo real?
No. Many AI tools leave no traces at all, and traces can be removed when a file is shared. Finding no traces only shows that the file has no readable trace.
Can I check a photo I downloaded from Instagram?
You can, but since Instagram often removes the file metadata and Content Credentials, the result will most likely be “No traces found.” Invisible watermarks may still be there, though, so try Google's SynthID tool as well.
Is there any point in checking a screenshot?
Very little. A screenshot is a new file and does not carry the original's file metadata or Content Credentials. Some invisible watermarks may survive in a screenshot.