AI Catfishing Went Industrial: How to Tell If a Dating Profile Is a Real Person

Your old playbook is not enough anymore

The standard advice still matters: reverse image search the photos, ask for a video call, check for a real social footprint. But the exposed operation showed why each check needs a backup.

• Reverse image search catches stolen photos, not generated faces. An AI-generated face has no original to trace back to. If a photo appears nowhere on the internet at all, that is not proof it is real. It is a different kind of warning sign.
• Video calls can be gamed. The operation reportedly staffed about 25 percent of its profiles with real gig workers who handled the things AI could not fake, including live video calls and social media follows. A face on a video call is still a strong signal, but it is no longer the final answer it used to be.
• People overestimate their ability to spot fakes. A 2026 threat report from Gen (the company behind Avast) found that AI was involved in tens of millions of video encounters worldwide in the first half of 2026, with roughly 1 in 30 flagged as scams, and that people are overconfident in their ability to detect deepfakes. Your gut is a worse detector than you think.

Signs you are talking to an AI persona

No single sign is proof. Look for several together.

The photos look almost real, but not quite. AI portraits keep improving, but they still slip on small details: lighting on the face that does not match the room, garbled text on a t-shirt or sign that looks like lettering but spells nothing, hands with the wrong number of fingers or fingers that merge together, teeth rendered as one fused white shape, hair that melts into the background instead of ending. Plastic, poreless skin on someone claiming to be an ordinary person in a casual snapshot is worth a second look.

The conversation is polished but empty. AI personas write smooth, generic copy and dodge specifics. Ask a question with a real answer: what street they grew up on, the bar they mean when they say “the usual spot,” what they actually did last weekend. A real person answers without thinking. A persona goes vague, changes the subject, or replies with a compliment that has nothing to do with what you asked.

They move fast and stay off the platform. The industrial operations share the classic scam playbook because it works: strong feelings in the first days, pet names instead of your name (easier to copy-paste to thousands of victims), and a fast push off the app to WhatsApp or Telegram where moderation cannot follow. Real romantic interest does not need to leave the platform on day one.

The story shifts. Small inconsistencies in basic facts, age, job, neighborhood, how long they have lived in the city. Humans misremember things, but a pattern of contradictions about a person’s own life is a flag.

What still works

Verification is now a habit of stacking weak signals, not a single knockout test.

• Ask for a live video call anyway. It remains the fastest screen, even though it is no longer decisive on its own. Add a small spontaneous ask, like holding up a specific number of fingers or answering a question about something visible in their room. Real people do this in five seconds. Someone stalling for weeks is telling you something.
• Check the social footprint with fresh eyes. A real account has history: old posts, tagged photos, comments from actual friends. A brand-new account with three posts is a flag. But remember that the exposed scam operation had its gig workers do social media follows precisely to pass this check, so treat it as one signal, not the verdict.
• Use mutuals. In a dense scene like NYC’s, you are rarely more than two degrees of separation from someone real. Ask who they know, where they go out, who else is in their circle. Specific answers are reassuring. Vague ones are not.
• Look for platform-level verification. Some platforms offer photo verification or identity checks as opt-in features, with badges on profiles that passed them. Seeking.com has said publicly that AI-assisted verification blocks more than 90 percent of attempted scammers before they send a message. That is the company’s own claim, but it points to where the real defense is moving: from user vigilance to platform systems. Verification tools reduce risk. None of them eliminates it.
• Treat “human-led” as a marketing claim, not a fact. The studio Anthropic exposed advertised itself as fully human-led while running mostly AI personas. A claim is not a verification system. Look for the actual tools: verification badges, human review, transparent moderation, real appeal paths.

The money rules

Every version of this scam ends at money, usually after weeks of trust-building: a medical emergency, a blocked bank account, an investment opportunity, crypto, gift cards. The rules are simple and they hold across every variant.

• Never send money, gift cards, or crypto to someone you have not met in person.
• Screenshot everything before you report and block.
• Report the profile to the platform, and report the fraud to the FTC at reportfraud.ftc.gov.

The FTC reported more than $1.16 billion in reported romance-scam losses in the first nine months of 2025. The scale is why the industrial operators keep coming back. Your money is the product. Do not deliver it.