How to Tell If a Photo or Video Is AI-Generated

Short answer: Start with provenance, not with your eyes: check for a C2PA / Content Credentials label, run a reverse image search to find an older original, and look at who published it first. Only then look for visual tells — garbled small text, warped hands and jewellery, physically impossible reflections, and lip-sync drift in video. No single tell is proof; two or three together usually are.

Detection tools that claim a percentage score are unreliable and get worse with every model release. Verification, on the other hand, does not age: an image that genuinely exists in the world usually leaves a trail — an earlier upload, a named photographer, a second angle from another person.

The six checks, in order

1. Look for Content Credentials (C2PA). Many cameras and generators now attach a signed history; social platforms surface it as an 'AI info' label.
2. Reverse image search the frame (Google Lens, TinEye, Bing Visual Search). An older, higher-resolution copy means the viral version is a derivative.
3. Trace the first poster. Brand-new accounts, no location, and a caption engineered for outrage are the strongest signal you have.
4. Zoom into small text: signage, badges, number plates and menus are still where generators fail most often.
5. Check physics: shadows pointing different ways, reflections that do not match the scene, ears and teeth that change between frames.
6. For video, mute it and watch the mouth, then listen without watching. Drift between the two is a common giveaway.

How reliable is each signal?

What to do before you share

Wait an hour. Genuine events get corroborated by other outlets and other camera angles within that window; synthetic ones tend to stay a single asset being re-uploaded. If you already shared something that turned out to be fake, delete and post the correction in the same place — corrections travel far less than the original, so speed matters.