AI Catfish: 12 Signs You're Talking to an AI-Generated Person

A reverse image search works by finding other copies of the same photo. A freshly generated face has no other copies, so it comes back clean, and a clean result feels like proof. It isn't. Treat "no matches found" as neutral information: it means the photo isn't recycled, not that the person is real. The stronger signal is whether the face appears anywhere with history attached, tagged by friends, in the background of someone else's post, in a years-old photo where they look younger.

Generated portraits are excellent at faces and still weak at everything around them. Look at the edges: earrings that don't match, ears with soft or missing detail, hair melting into the background, glasses whose frames don't line up behind the lens, teeth with wrong counts or fused edges. Look at context: text on signs or clothing that dissolves into pseudo-letters, straight lines that bend where they pass behind a shoulder, jewellery or watch faces that make no mechanical sense. Above all, look at the set as a whole. A real person's photos are inconsistent: different cameras, different years, different lighting, other people in frame. A generated set is suspiciously uniform, always flattering, always solo, always the same age.

Language models are fluent and shallow. They answer the question you asked and rarely carry a thread forward: you'll notice they never bring up something you mentioned three days ago unprompted, never ask a follow-up that only makes sense if they remembered your answer. Replies arrive with unnatural consistency, the same length, the same warm register, at any hour, with no typos and no "sorry, at work". Emotional language is generous but generic: you're "so different", "so easy to talk to", never "funny about your sister's dog". And specifics slide. Ask what street their gym is on, what their manager's first name is, what the weather was like this morning where they are, and you'll get pleasant vagueness or a confident answer that contradicts yesterday's.

Stop analysing and start requesting. Ask for a live video call, unscheduled, right now, and during it ask them to turn their head fully to the side, pass a hand slowly across their face, or hold up three fingers next to their cheek. Real-time face manipulation still breaks on occlusion and profile angles. Ask for a photo taken in the next two minutes holding a specific everyday object in a specific way, a spoon on their forehead is unglamorous and unfakeable at speed. Ask for a voice note answering a question you just invented. Any single refusal is nothing; a pattern of technical excuses covering every one of these requests is the answer.

AI catfishing is rarely done for fun. It's the entry stage of a funnel: build attachment cheaply at scale, then move the target to a private messaging app, then introduce money, an investment platform, a crisis, a gift card, a "verification" fee. Recognising the AI stage early is valuable precisely because it happens before any money is asked for. If you spot these tells, you are still upstream of the loss.