When you purchase through links on our site, we may earn an affiliate commission.Heres how it works.

Last week, I wrote about how frustrated I was thatChatGPTwouldnt display the textI wanted in an image.

In fact, they are too convincing.

Illustration of some binoculars.

A fake receipt created in ChatGPT.

ChatGPT is the first generally available, and free, AI image creator that can do convincing text.

People have already noticed thatChatGPT’s images are so good that you could use them tofake receipts.

But surely, there must be something we can do to spot the AI fakes.

An AI-generated receipt made using ChatGPT’s image generator

A fake receipt created in ChatGPT.

This is because of the massive increase in users since ChatGPT announced its new image-generating capabilities.

We added one million users in the last hour.

Its not the anime capabilities of ChatGPT that really intrigued me, though, its the improved text rendering.

Metadata window on a Mac.

Using Get Info on a Mac to reveal a file’s metadata.

Inside the metadata

Each image on your box or phone comes with metadata.

Images generated in ChatGPT are no different, and their metadata tells you everything you gotta know.

Images created using the ChatGPT website have ChatGPT.com in the Where from: part of their metadata.

Sightengine in a web browser

Our fake receipt easily fooled an online AI image checker.

In Windows, go to File Explorer locate the image then right-click it and selectProperties.

Click theDetailstab and scroll down to find the metadata that tells you what app it is from.

But the annoying thing is, even if you might see it, you might’t trust the metadata anyway.

It’s possible to edit this metadata.

It’s not easy to do, but if somebody is determined enough, they’ll find a way.

Who you gonna call?

So, if you’re able to’t trust the metadata, then what else can you do?

There are lots of online checkers that will tell you the likelihood of an image being AI-generated.

It said that it was most likely real, giving it just a 7% chance of being AI-generated.

Now that leaves me worried.

If the AI image checking websites can be fooled by the new ChatGPT images we’re all in trouble.

Right now I like to think I could still spot if an image is fake.

And of course it’s only going to get better over time.