How to Tell If an Image Is AI-Generated: 9 Tells to Check

AI images stopped being obviously fake somewhere in 2024. The six-fingered hands are mostly gone; the plastic skin got fixed. But "harder to spot" isn't "impossible to spot" — generators still leave fingerprints, and knowing where to look catches most fakes in a few seconds.

Here are the nine tells, ordered roughly from most to least reliable in 2026.

The Nine Tells

1. Text in the image

Still the most reliable giveaway. Signs, labels, book spines and background posters drift into almost-letters, inconsistent fonts, or words that dissolve halfway through. Zoom into any text before you trust a photo.

2. Edges where objects meet

Hair against backgrounds, glasses against faces, fingers holding objects. Generators blur, melt or invent geometry at boundaries — a fork that merges into a hand, earrings that fuse with hair.

3. Repetition in crowds and patterns

Background faces that are subtle copies of each other, brick and fabric patterns that tile unnaturally, identical leaves. Generators repeat themselves where they think you won't look.

4. Physics of light

Shadows falling in two directions, reflections that don't match the scene (check mirrors, windows, eyes), highlights with no light source. Lighting is coherent in reality and merely plausible in generations.

5. Jewellery, hands and teeth

Much improved, still imperfect: asymmetric earrings, rings that bleed into fingers, too many or too-even teeth, thumbs at odd angles. The classics still pay off, just less often.

6. Background logic

Staircases to nowhere, doors that couldn't open, furniture intersecting walls. Generators compose foregrounds carefully and improvise backgrounds — that's where the dream logic hides.

7. Uniform perfection

Real photos have noise: dust, smudges, uneven skin, motion blur on the moving thing only. An image where every surface is equally clean and every region equally sharp reads generated.

8. The context check

Not a visual tell, and stronger than most of them: where did this image come from? A reverse image search showing no original source, or a "news photo" that no news outlet carries, tells you more than any pixel.

9. Your unease

The weakest signal individually, but real — a composite of tiny wrongnesses you noticed subconsciously. Never treat it as proof. Treat it as the trigger to run checks 1–8.

The Honest Caveat

Top-tier generators now produce images where none of the visual tells appear. For anything consequential — money, news, reputation — provenance beats pixels: source, date, corroboration. The visual checks filter the 90% of fakes made with ordinary tools; the context check handles the rest. And AI-generated writing has its own tells entirely — covered in Is This Written by AI?

Knowing the Tells ≠ Having the Eye

Reading this list makes you slightly better. What makes you genuinely better is calibration — seeing hundreds of real-vs-AI pairs and being told when you're wrong. That's exactly what AI or Not does: a free daily game, 8 rounds a day, real photo or AI image (plus human-vs-machine text rounds). Streaks keep you honest; the reveal teaches you your blind spots.

Most people who think they have a good eye score worse than they expect on day one — and measurably better within a week. Play today's rounds and find out where you actually stand. Why this skill matters more every month: Everyone Needs an AI Radar Now.