Home / Learn / AI romance scams
Field note · Everyday AI
Published July 4, 2026 · Vita Indarra
Short answer: the romance long con — weeks of warmth, then an "investment opportunity" — is now run by AI at industrial scale: personalized from your own posts, patient because patience costs the scammer nothing, and running thousands of victims in parallel. You will not reliably spot the bot from its words. You end the scam with behavior: never send money or "investments" to someone you haven't met in person, tell one real person about any online relationship heading toward money, and treat every push for secrecy as the alarm itself.
Romance fraud always worked; it just didn't scale. A convincing long con needed a human to remember your dog's name, your shift schedule, your loneliness — for months, per victim. That labor was the scam's cost ceiling.
The ceiling is gone. Fraud researchers now describe operations where the conversation itself is automated: public profiles scraped to tailor the approach, language models maintaining warm, error-free, always-available chat with thousands of targets at once, and a human closer stepping in only when money gets near. The machine doesn't get tired, doesn't slip on details, and doesn't mind waiting three months. Patience — the thing that used to make a long con feel real — is now free.
Ten years ago, scam messages had tells: broken English, copy-paste blocks, forgotten details. Modern AI erased them. The writing is fluent, the memory perfect, the affection tuned to what you've shared. Honest advice in 2026 has to say it plainly: from text alone, you increasingly cannot tell. The tells that survive are behavioral — the platform exit, the never-quite-meeting, the 24/7 availability, and above all the slow bend toward money. Judge the pattern, not the prose.
Stop all further payments — especially any "fee" or "tax" required to withdraw; that demand is the tell, not a hurdle. Document the conversation, contact your bank or exchange immediately (early hours matter), and report to the FTC and the FBI's IC3. Then one warning that earns its own sentence: do not pay anyone who contacts you offering to recover your funds — "recovery agents" are a second scam industry that farms the victim lists of the first. Being targeted by a machine built to be loved isn't foolishness. Sending the second payment after the first went wrong is the only part you control.
Dating apps, "wrong number" texts, and social follows — increasingly personalized from whatever you've posted publicly. The opener is tailored because scraping your interests is free.
Party tricks ("ignore previous instructions") fail against hybrid operations where a human supervises. The behavioral pattern — platform exit, no meeting, money bend — is the test that works.
A real person who likes you will survive you verifying them; a scammer can't. Anyone who punishes caution with guilt ("don't you trust me?") has answered the question.
No money to anyone you haven't met in person. Ever. It ends the entire category.
Go deeper
The full anatomy of machine-made deception — voice clones, deepfake calls, automated cons, and the verification habits that don't expire, with the real, dated public cases — is in Don't Trust the Robot: the everyday reader's survival guide to AI you can't take at its word. No jargon, no doom — the habits that hold even as the fakes improve. Live on Amazon.