The 60-second check at a glance

Three steps, from coarse to fine. First, the origin: who posted the image first, and does the scene exist in other sources? Second, your own eyes: a few seconds of focused looking at the spots where AI images fail most often. Third, the technology: a forensic check that reads provenance data and image signals no eye can see.

The order matters because the steps cost different amounts of time and deliver a clear result at different rates. Step one alone often settles the case - say, when the reverse search shows that the "breaking storm photo" has been floating around the web for three years.

Step 1: clarify the origin (20 seconds)

Do not start with "does this look real?" but with "where does this come from?". An account posting under a real name for years is a different source than a two-week-old profile with number salad in the handle. Are reputable outlets reporting the supposed event? A spectacular photo that exists in exactly one post is a warning sign.

The strongest move in this step is the reverse image search: long-press or download the image and drop it into Google Lens, Bing or TinEye. That surfaces earlier publications, other resolutions and fact checks - and, as a bonus, exposes the classic trick of recycling a real but old photo in a new context.

Step 2: look closely (20 seconds)

AI image generators have become strong, but they still fail in predictable places. Check, in order: hands and fingers (count, posture, transitions), text in the image (signs, labels, tattoos - does it dissolve into fantasy characters?), ears, teeth and jewellery (asymmetries, merged earrings), shadows and reflections (do direction and content match?) and the background (bent lines, people with washed-out faces).

Then there is the overall impression: AI photos often look a touch too perfect - flawless skin, even light, a tidy scene. The full list with examples is in our guide to spotting AI images. But do not rely on your eyes alone: the best generators now produce images where even trained viewers get it wrong.

Step 3: get a technical check (20 seconds)

Now comes the part the eye cannot do. Upload the image to us or paste the link - the check runs automatically: we read provenance data like C2PA (cryptographically signed credentials that some generators and cameras embed), look for statistical AI traces in the image itself and let trained models assess the material. Shortly after, you get an assessment with a probability - and a shareable certificate in case you want to bring the result into a discussion.

One tip for better results: check as close to the original as possible. Screenshots and repeatedly forwarded versions lose metadata and image signals. How the check works in detail and where its limits are is covered in How does AI detection work?

When the result says "inconclusive"

Sometimes the evidence is not enough for a clear verdict - heavily compressed reposts are the most common reason. That result is not a bug, it is honesty: better an open "inconclusive" than a guessed percentage. What you can do then: get the original file (instead of a screenshot), look for a higher resolution, narrow down the context further via reverse search.

And until things are clear, the simplest rule of media literacy applies: when in doubt, do not share. An image that triggers strong emotions and cannot be verified is exactly the pattern disinformation runs on.