How to Tell if an Image Was Made With DALL-E or ChatGPT
OpenAI images are among the easier ones to check, because OpenAI marks them twice. It attaches Content Credentials to generated images and, since May 2026, an invisible SynthID watermark through a partnership with Google. Work from the original file, not a screenshot: the credentials are metadata and vanish easily, while the watermark is built to survive. When both are missing, fall back to a detector and the visual checks.
The two marks OpenAI adds
Content Credentials are signed C2PA metadata. OpenAI has added them to generated images since DALL-E 3 in February 2024, and is now a C2PA conforming generator, which means platforms can reliably read and pass along the manifest. It names the tool and the issuer, which makes it useful evidence.
SynthID is Google DeepMind’s watermark woven into the pixels. OpenAI applies it to images generated through ChatGPT, Codex, and the OpenAI API. Because it is in the image itself, it is designed to survive screenshots, resizing, and compression, the exact transforms that strip the credentials.
How to check them
Use OpenAI’s public verification tool at openai.com/verify, which checks both Content Credentials and SynthID for OpenAI-generated content. Note the scope: at launch it only covers OpenAI tools, so it cannot clear an image from another generator. You can also read the C2PA layer yourself at contentcredentials.org/verify, which works for any signed file.
What a screenshot does
A screenshot keeps the pixels and drops the metadata, so the Content Credentials disappear and the SynthID watermark remains. That is why provenance questions so often end at “we can’t tell.” When you can, ask for the original file. When you cannot, a detector is the next step.
Detector and visual checks as backup
Score the original file with the free AI image checker or the app and treat the result as a probability. Then run the visual checks for hands, text, reflections, and background logic, and combine the signals. Our honest breakdown of detector limits is in Do AI Image Detectors Actually Work?. If the image came from Google’s Gemini or Nano Banana instead, see the Gemini detection guide.
Common questions
Do DALL-E and ChatGPT images carry Content Credentials?
Yes. OpenAI has attached C2PA Content Credentials to images since DALL-E 3 in February 2024 and is now a C2PA conforming generator. The manifest names the tool and the issuer, but it lives in metadata and disappears in screenshots and many platform uploads.
What is SynthID, and do OpenAI images have it?
SynthID is Google DeepMind's invisible pixel watermark. Since May 2026 OpenAI has applied it to images generated through ChatGPT, Codex, and the OpenAI API, so the signal can survive transformations that strip metadata.
How do I check an OpenAI image for these marks?
Use OpenAI's public verification tool, which checks Content Credentials and SynthID for images generated with OpenAI tools. At launch it only covers OpenAI content. You can also read the C2PA layer yourself at contentcredentials.org/verify.
If there are no marks, is the image not from OpenAI?
Not necessarily. A screenshot or a re-upload can remove the C2PA manifest. The tool will not make a firm conclusion when it finds nothing, and neither should you. Treat missing marks as no evidence rather than proof.
Check any image in under 3 seconds
AI Detector scores an image and never stores it.
