How to tell if an image is AI generated

1. Start with the source, not the pixels

Before you zoom in on a single finger, find out who posted the image first and when. Photos of events that actually happened usually leave a trail: a photographer, an agency, other people's pictures of the same moment, earlier posts from the same account.

AI images often arrive fully formed, with no earlier copies, no other angles and nobody who will say they took them. That isn't proof on its own, but it tells you how much doubt to bring to the rest of the checks.

• Who posted it first? Is that account a person, a news outlet, or a page that posts a lot of striking pictures with no credits?
• Does the caption name a place, a date or a photographer you can check?
• Is anyone else showing the same scene from a different angle?

2. Look for the giveaways that still work

The famous tells (six fingers, melted text) are rarer with each new generator. When you do spot one it's strong evidence. When you don't, it means very little. Worth a look:

• Hands, teeth and ears: count, then check how they join up. Generators still fumble where things overlap.
• Text on signs, labels, screens and clothes: zoom in. Letters that turn into near-letters are a classic sign.
• Reflections and glasses: does the mirror, window or lens show what it should?
• Shadows: do they all fall the same way, and does each object have one?
• Jewellery, buttons and patterns: earrings that don't match, collars that merge into skin, stripes that change rhythm.
• Backgrounds: people who melt into crowds, railings that stop making sense, buildings without doors.
• Skin and hair: an even, airbrushed glow, or hair that dissolves into the background instead of ending in strands.

3. Check that the details agree with each other

Generators are good at making each part of an image look right and worse at making all the parts agree. Ask whether the picture is consistent with the story it is supposed to tell.

Does the weather match the date in the caption? Are the shop signs in the local language and script? Do the uniforms, number plates, power sockets and road markings belong in that country? Does the crowd look like a crowd, or like the same few faces repeated? A single mismatch is a reason to keep checking, not a verdict.

4. Run a reverse image search

A reverse image search looks for other copies of the picture. It can find an older post, the uncropped original, a creator's gallery where the image is labelled as AI art, or the same scene photographed by someone else.

Try more than one search engine, crop to the most distinctive part of the image, and pay attention to dates. If a picture of "today's flood" turns up in a post from three years ago, you have your answer before any detector gets involved.

5. Look for Content Credentials

Some AI tools, editors and cameras attach a signed record of how an image was made, using a standard called C2PA. You can read it with a free verify tool. If it says an AI tool generated or edited the image, that is about as direct as evidence gets.

Most images you meet online won't have one, because many social platforms and messaging apps strip this data on upload. So a missing credential proves nothing either way.

6. Use a detector as one more signal

A detector looks for statistical traces that generators and editors leave in the pixels, many of them invisible to people. Checkobot runs two: whole-image analysis on every picture, and face analysis on the largest face when it is at least 64 pixels across. It also reads Content Credentials. The result says AI or No AI detected, which detector fired, and why.

It refuses images under 256 pixels on the short side, and JPEGs compressed below roughly quality 50, instead of guessing. Upload the original file or paste a link to it. Screenshots and re-shares wash out the details a detector needs.

AI photo checker

7. Know where every method struggles

Eyes and detectors share some blind spots. Be extra careful with:

• Small AI edits inside a genuine photo: a removed object or a changed sign is harder to spot than a whole generated scene.
• Images from brand-new generators that no detector has seen yet.
• Screenshots, heavy compression and images forwarded many times.
• Illustrations, digital art and CGI, where "looks artificial" is the whole point.
• Strongly filtered or retouched photos, which can look AI-made to people and to models.

AI edited photo detector

Putting it together

No single check settles it. A clear source, earlier copies and consistent details make a strong case. A detector flag with no traceable source makes a strong case the other way. When the signals disagree, say so. "We couldn't confirm where this came from" is an honest answer, and a better one than a confident guess.

And keep any conclusion about the image, not the people in it. A flagged picture of a person doesn't make that person a fake.

Questions