A post circulating on X/Twitter claims that the image generation feature in ChatGPT-4o—the latest large language model (LLM) from OpenAI—can create realistic fake receipts, raising concerns about potential misuse for fraud.

The image shared in the post isn’t especially convincing at first glance. The text appears unnaturally superimposed on the paper and doesn’t bend with the wrinkles, which is often a giveaway of AI-generated visuals. However, we decided to explore the claim further.

A more compelling example came from a LinkedIn post by a French user, who reportedly asked ChatGPT to edit the text of a real restaurant receipt. The result was far more realistic, as it used an actual photo as the base—making the final output much harder to detect as fake.

To test the capabilities ourselves, we asked ChatGPT to generate a receipt for Black Dot Research. While the result had some inconsistencies—like an incorrect phone number and an inaccurate logo—it still looked fairly plausible at a glance. In the wrong hands, such images could be misused for filing fake reimbursement claims or inflating expense reports.

At the time of writing, we have not found verified cases of ChatGPT-generated receipts being used in actual fraud. However, the ease with which such fakes can now be produced suggests strong potential for misuse.

We rate the claim that fake receipts can be generated on ChatGPT and used for fraud as likely true.