Practical guide · verified against the real thing
Spotting AI fakes and deepfakes: the practical checks before you believe it
In one line: Generated images, cloned voices and fabricated text are now good enough to fool a first glance. Here are the concrete tells and the habits that stop you being taken in.
Generated media has crossed the line where a quick look is no longer enough — a convincing face, a familiar voice on a call, a plausible news quote. The defence is not paranoia or expertise; it is a few concrete checks and a habit of pausing before you believe or share something that was clearly designed to provoke a reaction.
The tells in images and video
Generated images still leak signs, though fewer each year: implausible hands and teeth, garbled text in the background, mismatched reflections or shadows, eerily smooth skin, and jewellery or glasses that melt into the face. Reverse-image-search a picture — if a "real" event photo exists nowhere else, treat it as suspect. And check the source before the pixels: a genuine news image appears on reputable outlets; a fabrication usually appears first on one account with no original.
The tells in voice and text
Cloned voice is the one that catches people, especially in a stressful "it's me, I need help" call. The defence is procedural, not perceptual: agree a safe word or call the person back on a known number, because a real person can verify and a scammer cannot. For text, be suspicious of writing that is fluent but factually hollow, or that arrives with unusual urgency — the same instincts as spotting a suspicious link, because AI is now used to make phishing more convincing, the risk covered in phishing training.
The habit that matters most
Slow down on anything engineered to make you act fast — outrage, fear, urgency, a too-good deal. Verify against a primary or independent source before sharing, because sharing is how fakes travel. None of this requires spotting every artefact; it requires refusing to let a convincing surface bypass the question "where did this actually come from?" That scepticism is the same one the desk applies to AI answers in why AI hallucinates — applied now to what you see and hear.
Next