AI or Not

AI or Not › Guides

Why AI image detectors are never 100% sure

If you have tried several AI image checkers, you may have noticed they sometimes disagree. That is not a bug; it is the nature of the problem.

AI tools keep getting better

Each new generation of image models fixes the mistakes the last one made. A clue that worked last year (like mangled hands) may not work this year.

Compression hides clues

Every time an image is screenshotted, resized or uploaded to a social app, it is compressed. Fine detail, noise and metadata that could help are lost. A screenshot of a screenshot is much harder to judge than the original file.

Real photos can look fake

Beauty filters, portrait mode blur, HDR and heavy editing all make real photos look smoother and stranger. Detectors can mistake these for AI.

Mixed images

Many images now are part real, part AI: a real photo with an AI-extended background, a removed object or a changed face. Those are the hardest to call either way.

How to use a probability

Treat the result as a weather forecast, not a court verdict. A 90% AI chance with clear clues is strong evidence. A 55% chance means "unclear". Combine it with context: where the image came from, who posted it, and whether the story checks out elsewhere.

That is why AI or Not shows its clues and a percentage. You see what it saw, and you decide.

Try it on your own picture

Is this picture real, or made by AI? Drop it in and find out.

Scan image

More guides

How to spot an AI-generated image

AI images on social media: how not to get fooled