| Field | What we found / documented |
|---|---|
| C2PA Content Credentials | Documented by OpenAI as included at generation time |
| Camera EXIF fields | Not applicable, no camera capture |
| Survives a screenshot or resave | No |
| Survives most social re-uploads | Usually no, platforms commonly strip it |
OpenAI's own documentation describes attaching C2PA metadata to images produced by DALL·E 3 in the API, in ChatGPT, and in Labs, so a file straight from those surfaces should register as having a Content Credentials manifest. That is a meaningfully stronger starting position than a generator with no stated policy: the credential is at least declared to exist at creation time, subject to independent verification of the signature and claims.
The practical catch is durability. A Content Credentials manifest is metadata, and metadata is routinely dropped by screenshots, most image editors' default "save for web" paths, and by many social platforms on upload, see our social platform metadata pages for specifics. If the copy you have was shared through a chat app or social feed, do not assume the credential is still attached; check it directly with our Content Credentials marker inspector, and remember that a marker's presence is a first step, not a validated signature.
Questions this page answers
Does an empty metadata result mean the image is not from ChatGPT / DALL·E?
No. A missing field can mean the metadata was never written, was stripped by an export or upload step, or was not inspected by this tool. Treat it as inconclusive, not as proof either way.
Does this page verify a specific image is from ChatGPT / DALL·E?
No. It describes documented or observed behavior for ChatGPT / DALL·E in general. To evaluate one file, inspect it with our metadata viewer and Content Credentials marker inspector and weigh the result alongside other context.