Let’s be honest—up until recently, most AI-generated images had one glaring flaw: text that made no sense.
But that’s changing fast. Thanks to upgrades in multimodal models like GPT-4 Turbo and DALL·E 3, AI can now generate text that actually looks real, complete with legible receipts, menus, and even ID cards.
Cool? Yes. Harmless? Not always.
Fake receipts are the latest “oops” use case for generative AI tools. A simple prompt in ChatGPT (especially with image generation enabled) can now whip up a fairly realistic-looking store or restaurant receipt. You can even ask it to copy an actual receipt, then tweak the prices or items as needed.
Throw in a little Photoshop magic—crease the paper, add a stain or shadow—and you’ve got something that could pass at a glance.
Table of Contents
How the Fake Receipts Work
Here’s how someone could do it (for educational awareness only, of course):
-
Upload a real receipt or describe it in detail.
-
Ask ChatGPT to recreate the receipt layout, line by line.
-
Edit prices, items, or dates directly in the prompt.
-
Generate the image.
-
Use tools like Photoshop to “rough it up” with textures.
They’re not perfect. Look closely and you’ll notice the numbers might not add up, or the fonts might be slightly off. But to the untrained eye? It’s believable enough.
Why This Could Be a Problem
Fake receipts aren’t new—people have been making them with basic design tools for years. But AI lowers the barrier to entry, making it almost effortless for anyone with internet access.
That opens the door to:
-
Inflated expense reports
-
Bogus refund requests
-
Fraudulent warranty claims
Even if 90% of people use it innocently—for art or fun—the 10% who don’t can cause real damage.
OpenAI’s Official Response
OpenAI isn’t panicking (yet). A spokesperson told TechPlugged:
“We monitor image generations on and off our platform… and take action when we identify violations of our usage policies.”
They also mentioned that C2PA metadata is embedded in AI-generated images—basically a digital tag that says, “Yes, this came from us.”
But here’s the catch: exporting or screenshotting an image without that metadata is a one-click job. So while safeguards exist, they’re easy to bypass.
Quick Wins: Spotting AI Fakes in the Wild
Worried about being fooled by a fake receipt? Here’s how to stay sharp:
- Check the math – totals and taxes often don’t add up correctly.
- Look for perfect alignment – AI tends to generate layouts that are too clean.
- Zoom in on text artifacts – look for blurry edges or uniform kerning.
- Request metadata – if it’s a digital file, ask for the original export.
And if you’re in doubt? Run the image through a C2PA metadata checker or AI detection tool like Hive.
