How to Spot AI-Generated Scam Messages, Photos and Videos
AI helps scammers write fluent messages, create fake faces and clone voices. What still gives AI-generated scams away, and why verifying through your own channel beats trying to spot fakes.
Artificial intelligence has made scams cheaper, faster and more convincing. Messages that once gave themselves away with poor spelling now read fluently in any language. Profile photos can show people who do not exist. Voices and faces can be imitated on calls. Some signs still help you spot AI-generated content, but the most important lesson is this: you do not need to spot the fake if you verify the request.
What AI changes, and what it does not
| AI makes it easier to | But the scam still needs you to |
|---|---|
| Write fluent, personalised messages in any language | Click, pay, or share a code or password |
| Create realistic photos of people who do not exist | Trust someone you have never verified |
| Clone a voice from a short recording | Act on an urgent call without calling back |
| Produce fake video of a public figure endorsing an investment | Send money to a platform you have not checked |
The tactics are unchanged. Our guide to the four stages of every scam applies just as much to AI-assisted scams.
AI-written messages
Signs that can still appear, though less and less:
- Oddly formal or generic wording that does not match how the sender usually writes.
- Perfect grammar from a "friend" who normally writes casually.
- Plausible but slightly wrong details, such as the wrong department name or an old job title.
Better test: ignore the writing and look at the request. Is it unexpected, urgent, and asking for money, codes or details?
AI-generated photos
Possible signs in fake profile photos:
- Earrings that do not match, glasses that merge into skin, or distorted hands and teeth.
- Blurred, warped or strange backgrounds.
- Every photo is a polished portrait, with no ordinary group photos or tagged pictures.
- No results in a reverse image search, because the face is new. See how to use reverse image search.
Cloned voices
A call that sounds exactly like your child, parent or boss can be generated from a short clip of their voice. Warning signs include:
- An emergency with a request for money, often by an unusual method.
- The caller cannot answer personal questions, or keeps the call very short.
- Background noise or a "bad line" used to explain odd sounds.
- A new number, or a reason why you cannot call them back.
Better test: hang up and call the person on the number you already have, or ask for your family code word.
Deepfake video
Fake videos of celebrities, business leaders or news presenters promoting investments are widespread in social media adverts. Fake live video calls are also possible. Possible signs include lips slightly out of sync, unnatural blinking or lighting, and a face that distorts when the person turns or puts a hand in front of it. But again, the stronger check is the request: no genuine celebrity endorses a guaranteed-return trading platform in an advert.
- A new number with an explanation of why the usual one cannot be used.
- Urgency, a payment to an unfamiliar account, and a request for secrecy.
- What to do: call your son on his usual number, or someone who is with him. Ask for the family code word if you have one.
Habits that beat any fake
- Verify through your own channel. Call back on a known number; open the official app yourself.
- Agree a family code word for emergencies. See our guide to setting up a family code word.
- Never share codes or move money because of an unexpected message or call.
- At work, require a second check for payment and bank detail changes, however senior the requester sounds.
- Limit public recordings of your voice and family details where practical.
Unsure about a message? Paste it or a screenshot into CyberWatch AI Scan for a free check.
AI changes the disguise, not the trick. For the full routine to identify and respond to scams, read our guide on how to identify and avoid online scams.
Frequently asked questions
Can scammers really clone a voice?
Yes. Voice cloning tools can imitate a person's voice from a short recording, such as a video posted online. That is why a familiar voice asking for urgent money should be checked by calling the person back on their known number.
Can I rely on spotting mistakes in deepfakes?
Not for long. The tools improve quickly, and many fakes are already hard to spot. Verification habits, such as calling back and using a family code word, work however good the fake is.
Are AI-written scam messages different from older scams?
The tricks are the same: urgency, fear, rewards and requests for money or codes. AI mainly removes the spelling mistakes and makes messages easier to personalise and translate.
Sources
- How to Recognize and Avoid Phishing Scams, US Federal Trade Commission
- How To Avoid a Scam, US Federal Trade Commission
- Phishing: spot and report scam emails, texts, websites and calls, UK National Cyber Security Centre
