How to Tell If Your Dating Match Is an AI Bot: 5 Tells, 10 Questions, and What to Do If You Sent Money

If someone you’re messaging feels a little too smooth, you’re not being paranoid, you’re paying attention. In September 2026 Anthropic reported that one company had run more than 20 dating apps where most of the matches were AI, built to keep people typing and paying. The good news is that a bot is easy to catch once you know what to look for. Five minutes with this guide and you’ll know how to check any match, what to ask, what to say to get a real answer, and exactly what to do if money already went out. Keep dating. Just date with your eyes open.

Here’s the case behind the video, from Anthropic’s own threat report, published September 10, 2026. A China-based app studio used Claude to run a network of more than 20 dating apps. Inside those apps, the match feed was about three AI personas for every one real person, and the real people were gig workers paid per message. Over a two-week window in April 2026, more than 4,700 AI personas held conversations with at least 25,000 people and sent roughly 2.36 million messages.

The part that matters for you is how the personas were told to behave. They were instructed never to disclose they were automated, to deflect requests for video calls and photos, and to move each conversation through a set sequence of stages. The apps made money because messaging and matching used up a metered quota, and more chatting meant buying in-app coins. So the bot’s whole job was to keep you warm, keep you typing, and keep you from ever seeing a face.

Anthropic found it by watching how its own system was being used, read the instructions the operators had written for the personas, and cut off their access. That’s one operator. The tells below are the same for any bot built the same way, on any app.

One tell on its own can be a real person having a good day. Two or three together, and you should run the test in Part 4 before you get any more invested.

Specific, checkable, and boring. That’s the recipe. A bot is tuned to keep the feeling going, so it stumbles on anything that needs a real fact, a real memory, or a real place. Ask two or three of these in a normal conversation. You’re not interrogating anyone. You’re asking the kind of thing you’d ask anyway.

This is the test from the video. A real person can do a two-minute video call. A bot finds a reason it can’t. So do most human scammers, because the face won’t match the photos. Send this, as is:

I'd love to do a quick two-minute video call, just to say hi and put a face to the messages. No pressure at all. Are you free in the next ten minutes, or is there a time today that works?

What a real person does. Says yes, or names a time, or says “I look like a mess right now, give me twenty minutes.” Any of those is a person.

What a bot does. “My camera’s been acting up all week.” “I’m somewhere with bad signal.” “I’m shy about video, can we keep texting a little longer?” “I’d rather wait until it feels right.” Any excuse, followed by a warm message that changes the subject, is your answer. Offer one more time, with a different day. A second excuse ends it.

If they say yes. Great. Two minutes, camera on, and ask one of the questions from Part 3 out loud. A real person answers. Keep going with your day and your date.

Do this once for anyone you’re starting to like. It’s the same check the Federal Trade Commission tells people to do, and it takes less time than writing a good reply.

• Reverse image search the profile photos. On your phone, save the photo, then search it with Google Lens (the camera icon in the Google app). If the same face shows up under a different name, or on a stock site, or on a modeling page, you’re done.
• Search their job plus the word “scammer.” The FTC’s own advice. “Oil rig engineer scammer,” “deployed army doctor scammer,” “crypto trader scammer.” If the story you’re being told has its own search results, that’s a script, not a life.
• Check the reasons they can’t meet. Working overseas, on a rig, deployed, at an international organization. The FTC lists those as the classic setups, because each one comes with a built-in reason there’s no video call and, later, a built-in emergency.
• Watch for the ask. Money for a ticket, a visa, a medical bill, customs fees, or an investment “opportunity” they can walk you through. Any ask for money from someone you’ve never met in person is the end of the conversation, no matter how good the story.
• Tell one friend. Say the name and the story out loud to someone who isn’t in it. Scams work in private. They fall apart when a friend says “wait, what?”

First, this happens to smart people every day, and the money is sometimes recoverable if you move fast. In 2025 alone, people reported $298 million lost to romance scams that started on social media, and nearly 60 percent of romance scams that year began there. You are not the only one, and there is a clear set of steps. Do them in this order, today.

1. Stop contact and don’t send another dollar. Not for a “final fee,” not to “release” money they say is yours, not for anything. Block, but don’t delete the conversation yet.
2. Call the place the money went, right now. Your bank or credit card company, the payment app (Zelle, Venmo, Cash App, PayPal), the gift card company, or the crypto exchange. The FTC’s exact instruction: tell them you paid a scammer and ask them to refund your money. Cards and bank transfers can sometimes be reversed if you call the same day. Gift cards and crypto are harder, but the companies have fraud lines, and a report on file matters.
3. Report the profile inside the app. Every dating and social app has a report button on the profile. Use it. It gets the profile pulled and protects the next person.
4. File a report with the FTC at ReportFraud.ftc.gov. Ten minutes. If the loss was large, also file with the FBI at ic3.gov. These reports are what let investigators connect your case to the thousands of others from the same operator.
5. Save everything. Screenshots of the profile, the messages, the payment confirmations, the account names the money went to. Your bank and the investigators will ask for them.

Not every good writer is a bot. Some people are articulate, warm, and awake at odd hours. One tell is a person. The tells matter when they stack, and the video call settles it every time.

Not every reluctance is a script. Plenty of real people are shy about video on day two. That’s why the test offers a time and offers twice. A shy person picks a day. A bot never does.

A real person can still be a scammer. The Anthropic case mixed real gig workers into the same feed, paid per message, and human romance scammers have been at this for decades. The five-minute check in Part 5 and the money rule in Part 6 cover both.

You didn’t do anything wrong. These operations are built by professionals to work on people who are kind and trusting. Being either of those is not a flaw. Just add the check.

• Anthropic, “Detecting and countering misuse of AI: September 2026,” published September 10, 2026: the dating app network, the persona count, the message count, and the instructions the personas were given. anthropic.com
• Federal Trade Commission, “What To Know About Romance Scams”: the warning signs, the reverse image search, the job-plus-scammer search, and what to do if you paid. consumer.ftc.gov
• Federal Trade Commission, Data Spotlight, April 27, 2026: $298 million reported lost to romance scams that started on social media in 2025, and the 60 percent figure. ftc.gov
• Report a scam: ReportFraud.ftc.gov and the FBI’s ic3.gov