How to Spot AI Generated Images

A few years ago, you could spot an AI image by counting the fingers. Too many? Almost certainly fake. Text on signs that looked garbled? Dead giveaway. These days, neither trick is reliable.

The major AI image generators, including Midjourney, DALL-E, and Adobe Firefly, have fixed most of the obvious errors that used to give them away. Hands look normal, text on signs usually reads clearly, and skin no longer has that waxy, oversmoothed look.

This matters because AI-generated images are not just a technical curiosity. They're turning up in AI investment scams, romance fraud, and political misinformation. Identity verification firm Sumsub has tracked a 94% year-on-year rise in deepfakeAI-made media that looks real but is actually fake. fraud attempts, with the UK among the most affected markets. Knowing what still works, and knowing what to do when something looks wrong, is now a practical skill.

Why the old visual tells no longer work

For most of 2022 and 2023, hands were the reliable test. Extra fingers, wrong angles, fingers that merged together. By 2026 that window has closed. The current versions of all three big generators draw anatomically correct hands nearly every time.

Text went the same way. Signs and labels now render cleanly. So did teeth, which used to come out as smooth white strips, and skin, which once had that painted-on quality. None of the three tells you anything useful any more.

Ask people whether they can spot an AI image and most say yes. Put them in front of a test and the hit rate collapses. That gap between how good we think we are and how good we actually are is the thing scammers are selling into.

What to look for: 7 clues that still hold up

What follows is harder for a generator to fix, because these are failures of physics and internal consistency rather than of rendering quality.

1. Cloned faces in crowds. AI models fill background scenes by recycling facial patterns. Look at people behind the main subject. Do two or three of them share suspiciously similar features, the same nose shape, the same jaw? Real crowds are random. AI crowds sometimes repeat.

2. Floating edges where hair meets backgrounds. Hair is computationally expensive to render accurately. Where individual strands meet an outdoor sky or a busy interior, look for a blurry halo or an unnaturally smooth edge. This catches AI images regularly, even in 2026.

3. Shadows that contradict the light source. In a real photograph, every shadow points in the same direction. AI images sometimes give different objects their own imaginary light source. Check shadows on the face, the floor, and any objects in the scene. Do they all agree about where the light is coming from?

4. Reflections with no logical physics. Puddles, windows, glasses lenses, and polished surfaces should reflect their surroundings accurately. AI often gets this wrong, producing a smeared blur or the wrong scene entirely where a clear reflection should be.

5. Depth of field that looks painted on. Real camera lenses blur backgrounds according to optical physics. AI images sometimes blur backgrounds unevenly, or blur a foreground object that should be in sharp focus. It looks less like a photograph and more like a filter applied in the wrong direction.

6. Background elements that collapse under scrutiny. A pub sign with almost-English text. A bookshelf with spines that merge into each other. At thumbnail size these details look fine. Zoom in and they fall apart.

7. Subjects sharing features they shouldn't. This is the one that stays with you. Two unrelated people in the same photo turn out to have identical ear shapes, or three strangers in an "office crowd" share a jawline. Generators copy features between people in a single image, and once you have noticed it happening you cannot stop noticing it.

Should you use an AI image detector?

Yes, though with some caveats. A few tools are genuinely worth knowing about.

DeepAI Image Detector is free and takes seconds to use. Upload an image or paste a URL and it gives you a probability score. Treat it as a promptThe message or instruction you give an AI. for further investigation, not a final verdict.

Content Credentials, backed by a coalition of technology companies and camera manufacturers through the Coalition for Content Provenance and Authenticity (C2PA), lets you paste an image URL at contentcredentials.org/verify to see its creation history. If an image carries a credentials badge, you can see exactly how and where it was made and edited. No credentials doesn't mean it's fake, but credentials present do meaningfully increase trust.

Google SynthID embeds an invisible watermark into images produced by Google's own AI tools. It survives cropping, compression, and screenshots. Platforms and services that integrate with Google's verification system can confirm whether an image came from a Google AI tool. It only covers images made with Google's own generators, but it adds a layer of provenance that other watermarks don't provide.

They all share the same weakness. Detection tools do well against older generators and struggle badly against the newest commercial models, where accuracy drops off a cliff. Use them alongside your own eyes and ordinary common sense, never instead of them.

Why this matters for UK scams

AI-generated images are showing up in specific scam types that UK adults are encountering on social media and WhatsApp.

Celebrity investment deepfakes. Fake images and short videos of Martin Lewis, Eamonn Holmes, and other recognisable UK figures have been shared on Facebook and X to promote fraudulent investment schemes. They look real enough to be convincing at a glance. If you see a well-known British person endorsing an investment opportunity you haven't heard of, treat it with suspicion. Our guide on AI investment scams in the UK covers this in detail.

Romance fraud with AI-generated profile photos. Scammers use AI-generated portraits to create online identities that look real. The photos are plausible and often attractive, but the person doesn't exist. A reverse image search used to help catch these. Today's AI faces often pass that test because the images are unique, not copied from elsewhere. We've covered this fully in our guide on AI romance scams.

Proof-of-life manipulation. Some scammers use AI-generated images to falsely "prove" that a family member is safe, or in danger, to pressure someone for money. If an image is being used to put emotional pressure on you, be sceptical regardless of how convincing it looks.

What UK law says about fake images

Creating and sharing non-consensual fake intimate images are criminal offences in England and Wales under the Online Safety Act 2023 and related legislation. Possessing them with intent to share is also an offence.

Using AI-generated images to commit fraud is prosecuted under existing UK fraud legislation. Report it online at reportfraud.police.uk, the national online fraud reporting service.

The verification approach: smarter than inspection

Looking hard at an image will only get you so far. Working out where it came from gets you further. Before sharing or believing a surprising image, run through these questions.

• Who posted this first? A dramatic image that exists only on one anonymous account, with no news coverage anywhere, deserves scepticism regardless of how realistic it looks.
• Can a reverse image search trace it? Paste the image into Google Images or TinEye. If it appeared from nowhere recently, or the same image shows up in completely different contexts, be wary.
• Does it carry credentials? Try contentcredentials.org/verify. If the image has a Content Credentials badge, you can see its full creation history.
• Is it being reported by reliable sources? If a supposedly shocking image is spreading widely, check Full Fact or BBC Verify before passing it on.

The most useful habit is a simple pause. Ask yourself: who benefits from me believing this? Where did it actually come from? Can I find confirmation from a news source I trust?

What to do if you spot a fake

If you see an AI-generated image being used to deceive or defraud:

• Do not share it, even to call it out. Sharing spreads it further.
• Report it to the social media platform using their in-app reporting tools
• If you've been targeted financially, call your bank immediately
• Then report it at reportfraud.police.uk

For a full guide to every reporting route available in the UK, see our article on how to report an AI scam. For voice-based fakes, calls pretending to be family members or bank staff, see our guide on AI voice scams for what to listen for.

If you'd like to build your confidence with AI tools safely and at your own pace, the WellWired online course covers exactly this: spotting fakes, using AI confidently, and knowing what not to share.

Frequently asked questions

Can I tell if a photo is AI generated just by looking?

Sometimes, but not reliably in 2026. Studies find people correctly identify high-quality AI-generated images only around a quarter of the time, even those who consider themselves tech-savvy. Visual inspection is a useful first step, not a final answer.

Is there a free tool to detect AI images?

Yes. DeepAI's image detector (deepai.org/ai-image-detector) is free and works in seconds. Google's SynthID detector is built into Chrome; right-click any image and choose "About this image." Content Credentials (contentcredentials.org/verify) is also free and shows an image's full creation history if it carries a credentials badge.

Can AI detectors be wrong?

Yes, regularly. Current tools perform reasonably against older AI generators but accuracy drops sharply against the newest commercial systems. Never rely solely on a detector's verdict. Combine it with a reverse image search and consider the context of where the image appeared.

Are AI-generated images illegal in the UK?

Using AI-generated images to defraud someone is a criminal offence under UK fraud law. Creating and sharing non-consensual fake intimate images are criminal offences under the Online Safety Act 2023 and related UK legislation. Possessing them with intent to share is also an offence.

What should I do if someone sends me a suspicious image?

Do not share it. Run a reverse image search on it. Try contentcredentials.org/verify to check its history. If the image is being used to pressure you for money or personal information, treat it as a scam. Contact your bank and report it to Report Fraud at reportfraud.police.uk.