Why do AI detectors flag real photos?
A detector estimates how closely an image resembles the AI images it learned from. It does not read where the image came from, so real photos are sometimes scored as AI. In a May 2026 NewsGuard test of 15 authentic news photos, five detectors called the real images AI 13.33 percent of the time overall: ScamAI flagged 6 of the 15, ZeroGPT 3 and AI or Not 1, while Hive and Sightengine flagged none. ZeroGPT’s chief executive told NewsGuard that resizing and compression can lead to real images being flagged.
A lower score from a second detector is still only an estimate, so it rarely ends the argument. The optional score on this site comes from Sightengine, one of the five tools in that test, and is one signal, not proof.
Source: NewsGuard: AI image detectors call authentic photos fake (May 2026) (opens in a new tab)
How do I prove my photo is not AI?
Start with the original file. The copy someone ran through a detector is often a screenshot, a download from a social app or a resized export, and those have usually lost the records that show where a photo came from. Collect the file as your camera or phone saved it, then add what an AI generator does not leave behind.
- Phone: copy the photo straight from the phone’s photo library or camera folder, not from a chat app or a social post. In Photos on a Mac, File > Export > Export Unmodified Original keeps the file as it was captured.
- Camera: use the file on the memory card, and the RAW file too if you shot RAW and JPEG.
- Other frames from the same moment, shot just before or after, with matching capture times.
- Anything that places you there: the location, the people with you, other photos from that day.
Check what your original file declares
Choose the original in the checker above. It reads the file in your browser tab and does not upload it. These are the results you are most likely to see:
- No provenance signals found, with EXIF metadata parsed: the usual result for a phone or camera photo. The file declares no AI generation. That supports your account but does not prove the photo is real on its own.
- Content Credentials found and valid: some cameras sign each photo. Google says Pixel Camera on the Pixel 10 attaches Content Credentials to every JPEG it takes, and the Leica M11-P signs photos when its Content Credentials setting is on. A valid state means the file still matches what was signed. The report names the signer only for a few companies, Google among them; for other camera makers it shows the credential and its validation state.
- Content Credentials found, but they did not validate: the file was probably changed after signing, or its certificate could not be checked. Send the file as it came out of the camera instead.
- AI generation declared: the file records an AI step, such as a generative edit in Photoshop or Google Photos. Say so plainly and explain what the AI tool changed; leaving it out would undercut everything else you send.
Source: Google: Content Credentials in Pixel 10 and Google Photos (opens in a new tab)
Show the camera details
The report lists which metadata sections a file contains but does not display fields such as camera model, lens or capture time. To show those, open the photo’s information panel: on an iPhone swipe up on the photo or tap the info button, on Windows open the file’s Properties and then Details, and on a Mac use Preview’s Tools > Show Inspector. Anyone can edit these fields, so present them alongside the original file rather than instead of it.
Can I prove my art is not AI-generated?
For a drawing or painting, the process is the strongest evidence. Keep the layered working file, such as a PSD or Procreate file, along with sketches and saved versions. Procreate records a time-lapse of every new canvas by default, and you can export it from Actions > Video. For physical work, photos of the piece at different stages, taken over several days, do the same job. Our guide Is this art AI? covers what reviewers look for.
Source: Procreate Handbook: time-lapse video (opens in a new tab)
What to send the person who flagged it
Send evidence they can check, and ask for the decision to rest on it rather than on one score. Our guide to detector accuracy explains why a single score is a weak basis for it.
- The original file and the other frames or process files, not a screenshot.
- What the checker found, in your own words or as a screenshot of the report.
- What you did to the image after taking or making it, including any AI tools.
- A request to know which detector was used, what score counted as AI, and whether a person will look at your evidence.
What not to do
Do not edit the metadata, add a camera name or run the image through a tool that claims to make it pass a detector. A changed file fails the very checks that support you, and it looks like tampering. Image Evidence does not offer metadata editing or detector evasion.
What this evidence cannot prove
A valid credential shows that a file has not changed since a camera or app signed it; it does not prove that what the photo shows happened as described. Missing credentials prove nothing either way, because most cameras do not write them. The checker does not detect SynthID or Meta Content Seal watermarks, and it does not analyze pixels unless you ask for the optional score. The decision belongs to the person reviewing your evidence.