You are scrolling over breakfast when a photo stops you: a famous landmark half underwater after a storm, shared thousands of times. It looks real, but so do many pictures that software made in seconds. Knowing how to tell if an image is AI generated is now an everyday skill, and WhatsOnTech keeps it practical: the clues worth a glance, the checks that work better, and what none of them can prove.
Why a picture is worth a second look
Plenty of AI images are harmless fun: a joke, a fantasy landscape, a birthday card. But image generators, close cousins of the AI tools that draft your emails, can also be used to mislead. The misleading ones tend to turn up in four places:
- Scams. A celebrity seeming to back an investment, or a fake charity appeal after a disaster.
- Fake news. A dramatic scene that never happened, shared fast because it confirms what people feel.
- Fake products. An advert for a flawless item that may not exist, or looks nothing like what arrives.
- Fake profiles. A friendly face on a dating app or social account that belongs to nobody.
WhatsOnTech’s rule of thumb: the more a picture wants from you, whether money, a share or your trust, the more checking it deserves. A cute dog in a hat can stay a cute dog in a hat.
Visual clues that sometimes give it away
AI image tools build pictures from learned patterns, not real scenes, so they sometimes slip on details a camera would simply record. Zoom in on:
- Hands and fingers. Extra, missing or fused fingers, or a grip at an impossible angle.
- Text and signs. Shop signs, labels and number plates that turn into near-letters up close.
- Jewellery and ears. Mismatched earrings, glasses melting into skin, oddly shaped ears.
- Lighting and shadows. Shadows falling in different directions, or a face lit from nowhere.
- Backgrounds that melt. Railings, windows and crowds that bend, repeat or blur together.
- Too-perfect skin. A glossy, poreless look, like a magazine cover with the texture sanded off.
The honest catch: these clues fade with every new version of the tools. Newer images often get hands and text right, while real photos can have odd shadows or heavy filters. The WhatsOnTech take: treat a clue as a reason to check further, never as proof.
How to tell if an image is AI generated from its source
Pixels can fool you, but a picture’s history is harder to fake, which is why WhatsOnTech suggests starting with where it came from. Ask who posted it originally: a week-old account with no other photos deserves less trust than a named photographer or an established organisation. Then check whether a trusted outlet shows the same thing, since real storms and protests get photographed from many angles.
A familiar name is no guarantee either. A friend’s hacked account can share fake giveaways, one more reason for keeping your own accounts locked down.
Reverse image search in a minute
A reverse image search looks for the picture itself, not words, and shows where else it appears online. With Google’s search with an image feature, you upload a file or, in Chrome, right-click an image and choose Search with Google Lens. TinEye, a free site, does a similar job.
The results might reveal an older copy with a different caption, a fact-check, or an artist who first posted it as AI work. A “breaking news” photo that appeared online years ago is not breaking news.
Content Credentials and AI labels
Some cameras, editing apps and AI tools can attach Content Credentials, a provenance record (a note of where a file came from and how it was edited). They are built on C2PA, an open industry standard, and can show up as a small “CR” pin, which the Content Credentials website explains. Some social apps add their own AI labels too.
Both help when present, but neither is complete. Credentials can be lost when an image is screenshotted or re-uploaded, and platform labels rely partly on uploaders being honest. So take an AI label seriously, but a missing label proves nothing, and online “AI detector” scores can be wrong in both directions.
Video and voice deepfakes
A deepfake is a video or audio clip where software swaps a face or copies a voice, so someone appears to say things they never said. The same checks apply: find the source, the original and any trusted reports. Out-of-sync lips or a flat, oddly paced voice can be hints, but they are fading too.
The extra step is to verify through a second channel. If a video call from your “boss” asks for an urgent payment, or a relative’s voice pleads for money, hang up and call them on a number you already have. The FTC, the US consumer protection agency, gives the same advice about AI-cloned voices in family emergency scams: do not trust the voice, check the story.
Requests like these follow the pressure pattern behind most scams, whatever face or voice delivers them. In WhatsOnTech’s view, the request matters more than the messenger.
A two-minute checklist
Here is how to tell if an image is AI generated when it matters, most useful checks first:
- Pause before you share or pay. A strong emotional reaction is your cue to slow down.
- Find who posted it first. Then look for a trusted second source.
- Run a reverse image search. Look for older copies, other captions and fact-checks.
- Check for labels. A “CR” pin or an AI label helps; a missing one proves nothing.
- Zoom in on the details. Hands, text, ears and shadows, though clues can mislead.
- Verify people another way. Call back on a number you already know.
The takeaway
How to tell if an image is AI generated comes down less to counting fingers and more to asking where a picture came from and what it wants from you. Often you cannot be completely sure, and that is fine: judge the claim, not just the picture. A real event leaves a trail of sources, while a fake often leaves a single viral post. Checking the story before the pixels is a habit worth keeping, and WhatsOnTech’s AI tools guides have more like it.