Labs · Test pending
What metadata does a ChatGPT-generated image contain?
Short answer
We have not published a dated dump of a ChatGPT-downloaded image yet. Possible file-level signals include ordinary metadata and, when present, C2PA / Content Credentials. A missing signal is inconclusive — not proof the picture is human-made. Screenshots throw away container metadata.
Test pending
- Published
- Last updated
- Last tested
- —
What we measured
No dated CreatorPrivacyKit result is published for this route yet. Fields below stay empty on purpose.
| Field | Delivered copy |
|---|---|
| GPS / location | Not yet tested |
| EXIF | Not yet tested |
| Camera information | Not yet tested |
| Timestamps | Not yet tested |
| Provenance / C2PA | Not yet tested |
| Recompression / resize | Not yet tested |
GPS / location
Not yet tested
EXIF
Not yet tested
Test environment
Environment details are recorded when a test is run: OS, app or browser version, device, and upload path.
How to test your own file
- Download the image from ChatGPT, do not screenshot.
- Inspect C2PA, EXIF, XMP, PNG text locally.
- Optionally run official verification where a vendor provides it.
- Record model/product UI version.
Methodology
Every published result has to follow the same protocol: a known source file, a recorded hash, a specific upload path, and a downloaded copy. Folklore does not count.
Privacy implications
If you need to inspect origin claims, use the original download, not a screenshot. Do not advertise a detector as 100% accurate.