How to spot AI-generated scams, fakes and dodgy emails
AI has made scams cheaper, faster and far more convincing. Here is how to spot the fakes and protect your business.
For years we taught people to spot a scam by its bad spelling and clumsy English. That advice is now dangerous, because AI writes flawless, convincing English for free. The polished email, the perfect-looking invoice, the message that sounds exactly like a colleague: the old tells are gone, and a lot of people have not updated their instincts.
This is not a reason to panic, it is a reason to change what you look for. The good news is that the new defences are about behaviour and verification, not spotting typos, and they are easy to learn.
Stop trusting polish
The first thing to unlearn: a well-written, professional-looking message is no longer a sign of legitimacy. Scammers now sound articulate, on-brand and urgent. Judge the request, not the writing. A perfect email asking you to do something unusual is more suspicious now, not less.
Watch for the real tell: urgency plus an unusual action
Almost every scam, however polished, needs you to do something out of the ordinary, quickly. Pay a new account. Change bank details. Buy gift cards. Click to "reactivate" something. Move money to be safe. The combination of pressure and an unusual money-or-access request is the signal that survives AI. Slow down exactly when you are being rushed.
The pattern to fear: a believable message, a sense of urgency, and a request to move money, change details, or hand over access. Treat that combination as a stop sign, regardless of how real it looks.
Verify on a separate channel
If your "supplier" emails new bank details or your "boss" messages asking for an urgent payment, confirm it another way, a known phone number, in person, a fresh message you start yourself. Never reply on the same thread or call a number the message gives you. This one habit defeats the vast majority of these scams, AI-made or not.
For years, fluent writing worked as a signal of legitimacy because producing it took a real person with real effort. Generative models broke that link. A large language model, the technology behind tools like ChatGPT, drafts flawless, on-brand English in any tone for nothing, so polish no longer tells you anything about who or what is on the other end. The same shift has hit audio and video. A deepfake, a synthetic clip generated by a model trained to mimic a specific person, can now clone a recognisable voice from a few seconds of sound and a convincing face from a handful of images, both scraped easily from social media. The uncomfortable takeaway is that every channel which used to carry a built-in authenticity cue, the way someone writes, the sound of their voice, their face on a call, can now be imitated cheaply. That is why the only reliable defence left is out-of-band verification, confirming through a separate channel you chose, not the one the message arrived on.
Know that voices and faces can be faked too
It is now possible to fake a familiar voice on a phone call or a face on a video from very little. If a call or clip feels off, or makes an unusual request, the fact that it sounds or looks like someone you know is no longer proof. Verify the same way: a separate, trusted channel.
Make it a team rule, not a personal worry
The businesses that get caught usually had no agreed process. Set a simple rule everyone follows: any change to payment details or any urgent money request gets verified on a second channel, no exceptions, no embarrassment for asking. A boring rule, applied every time, beats clever instincts that have an off day.
AI has made the fakes better. It has not changed the fundamentals: pressure plus an unusual request equals stop and check. Hold onto that and you are far harder to fool than most.
If you want your team set up with simple, solid defences against this new wave of scams, that kind of practical safeguarding is something we can help you put in place.
Book a quick chat →Related: How to use AI at work without leaking your data.
Common questions
How do I spot an AI-generated scam?
Stop relying on bad spelling, AI writes flawlessly now. Watch instead for the real tell: urgency combined with an unusual request, like changing bank details or moving money. Judge the request, not the polish.
Can AI fake someone's voice or face?
Yes, convincingly, from very little source material. If a call or video makes an unusual request, sounding or looking like someone you know is no longer proof. Verify on a separate, trusted channel before acting.
What is the best protection against AI scams?
A simple team rule: any change to payment details or any urgent money request gets confirmed on a second channel you trust, every time, no exceptions. A boring rule applied consistently beats instinct that can have an off day.