Was It Made by ChatGPT or Gemini? How to Check an Image

ChatGPT and Google's Gemini are two of the most-used image generators, and both now label what they make. That means you can often check an image's origin for certain, as long as you use the right tool and the label hasn't been stripped. Here is how each label works and how to check it.

Two kinds of label

• Content Credentials (C2PA): a signed record inside the file that says which tool made it and that AI was used. Any C2PA reader can check it, but screenshots, re-saves and many apps remove it.
• SynthID: an invisible watermark woven into the pixels, created by Google DeepMind. It is designed to survive common changes like compression and screenshots, but only the maker's own tools can check for it.

Images from ChatGPT (OpenAI)

OpenAI has added Content Credentials to images from its image tools for some time. In May 2026 it announced it would also add the SynthID watermark, and it launched a public checker at openai.com/verify that looks for both signals in images from ChatGPT, the OpenAI API and Codex.

• To check the Content Credentials: add the image to the OriginLens Image Analyzer. If the record is there, the File labels card says the file declares AI generation, and the Provenance tab shows OpenAI as the tool.
• To check the watermark: use OpenAI's own verifier. Watermarks can't be read by other tools.

Images from Gemini and other Google AI

Images made or edited with Google's AI models carry a SynthID watermark. Google has also started adding Content Credentials to images from its newer image models (the Nano Banana family) in the Gemini app and some of its other products.

• To check the watermark: open the Gemini app or gemini.google.com, attach the image, and ask "Was this created or edited with Google AI?". If Gemini finds SynthID, all or part of the image was made or edited with Google AI. Use one image at a time; if you use a screenshot, crop it tightly.
• To check the Content Credentials: add the image to the OriginLens Image Analyzer. Google often records the AI step in an earlier layer of the record (for example, "created with Google Generative AI", then a watermark added on top). OriginLens reads every layer, so it still finds the AI label.

A real example: a portrait edited with Google's AI that we tested had two signed layers from Google. The latest said only "added visible watermark"; the earlier one recorded the AI step as trainedAlgorithmicMedia, the industry term for AI-generated media. A checker that reads only the latest layer would have missed it.

Which check to use

What a "no" really means

These checks are reliable when they find something. When they find nothing, they prove much less:

• OpenAI's checker can only tell you about OpenAI's images, and Gemini only about Google's.
• Images from other generators (such as Midjourney or open-source Stable Diffusion models) usually carry neither label.
• Screenshots and social apps often strip Content Credentials. Watermarks hold up better, but heavy editing can still weaken them.

So "no label found" means "unknown", not "real". When that happens, try a reverse image search to find the original, and read our guide on how to tell if an image is AI-generated.

Step by step

1. Get the original file if you can, not a screenshot.
2. Add it to the OriginLens Image Analyzer and read the File labels card and the Provenance tab.
3. If you suspect ChatGPT, check it at openai.com/verify. If you suspect Gemini or Google AI, ask the Gemini app.
4. If nothing is found, run a reverse image search and treat the result as unknown.

Frequently asked questions

Can Gemini tell if an image was made by ChatGPT?

Not reliably. Gemini's watermark check only recognises Google's SynthID. Use a Content Credentials reader such as OriginLens, or OpenAI's own checker.

Does a screenshot remove these labels?

A screenshot removes Content Credentials, because it creates a new file. A SynthID watermark is designed to survive screenshots, so the maker's checker may still find it.

Is there one tool that checks everything?

Not yet. Content Credentials are an open standard any tool can read, but each company's watermark can only be checked with its own tool. See our comparison of free AI image checkers.