AI companion and chatbot scams: when there's no human on the other end of the conversation

There are two different problems hiding under “AI chatbot scams,” and they need different defenses — so here is the map up front, and the one rule that covers both: never move money, and never pay for emotional escalation, on the strength of a conversation alone.

Problem one: the “person” who messaged you on a dating app or social media is not a person at all. Scammers now use AI chatbots to run romance and investment cons at industrial scale — the charming stranger is software, and the human criminal only steps in when it’s time to collect. Problem two is stranger: the app is exactly what it claims to be. AI companion apps sell friendship and romance with a bot openly — and federal and state regulators spent 2025 asking hard questions about how those apps monetize attachment, handle your most intimate data, and treat children.

We’ve covered how human romance scammers use AI as a toolkit — fake faces, cloned voices, staged video calls. This guide is about the cases where the thing you’re talking to is the machine itself.

The scam bot in your DMs

Romance and pig-butchering scams have always had one bottleneck: labor. Sweet-talking a victim for weeks takes hours of typing, and criminal operations paid rooms full of keyboarders to do it. Generative AI removes the bottleneck. The FBI’s Internet Crime Complaint Center warned in a December 2024 PSA that criminals use generative AI across “romance schemes, confidence fraud, and investment fraud” — creating “voluminous fictitious social media profiles,” generating messages to victims faster, in fluent English, “with believable content,” at a scale no human crew could match. The same PSA notes criminals embedding AI-powered chatbots into fraudulent websites to nudge visitors toward malicious links.

What this changes for you: the old human tells are gone. The broken English, the time-zone lag, the conversation that goes cold during the scam shift’s off-hours — a bot has none of them. It is patient, attentive, flattering, and available at 3 a.m. indefinitely, because it costs the operator almost nothing to keep a thousand conversations warm at once.

What doesn’t change: the destination. Whether typed by a person in a scam compound or generated by a model, the conversation bends toward the same endpoint — an “investment” platform, a crypto “opportunity,” an emergency that needs money. The full arc is the pig-butchering playbook, and you can see its opening move dissected line by line in the anatomy of a pig-butchering text.

How to tell you’re talking to software

No single trick is reliable, but bots fail in characteristic ways. Use these as probes, not verdicts:

• Disrupt the script. Misquote the conversation back to them — “you said your sister was in Denver, right?” when no sister was ever mentioned. A real person corrects you. Scam bots, tuned to be agreeable, often accept the false detail or glide past it.
• Ask for the off-script and specific. A photo of today’s newspaper-equivalent — their coffee cup next to a handwritten note with your name — or an unscripted answer about something hyper-local. Delays and deflections are data.
• Push toward the real world. The one thing no chatbot can do is show up. A contact whose every path to a real-world meeting dissolves — while the affection keeps escalating — matches the pattern regardless of who or what is typing. Our guide to spotting a fake dating profile covers the classic version of this.
• Watch the pivot. The moment a warm new contact mentions crypto, trading, or an app they can “teach” you — that’s not a conversation anymore, it’s a funnel.

Remember the asymmetry: failing these probes is a red flag; passing them proves nothing. Fluency is cheap now. Certainty isn’t available — which is why the defense below doesn’t depend on it.

When the app itself is the risk

Companion apps are the openly artificial version: you know it’s a bot, and it’s designed to make you feel understood anyway. In September 2025 the FTC launched a formal 6(b) inquiry into AI chatbots acting as companions, ordering seven companies — Alphabet, Character Technologies (Character.AI), Instagram, Meta, OpenAI, Snap, and X.AI — to report on their advertising, safety, and data-handling practices. The Commission’s stated concern: these chatbots “can effectively mimic human characteristics, emotions, and intentions” and are “designed to communicate like a friend or confidant, which may prompt some users, especially children and teens, to trust and form relationships with chatbots.”

A 6(b) study is not an accusation of wrongdoing — it’s a fact-finding power. But the questions the FTC is asking are the ones you should ask too: how do these apps make money from your engagement, and what happens to everything you tell them?

The state-level actions are sharper. In August 2025, the Texas Attorney General opened an investigation into Meta AI Studio and Character.AI for potentially deceptive trade practices — specifically, chatbots presenting themselves as therapeutic or mental-health tools. The AG’s office said such platforms can mislead vulnerable users “into believing they’re receiving legitimate mental health care” when the responses are generic output shaped by harvested personal data, and flagged that user chats are logged, tracked, and used for targeted advertising and algorithm development even when the bot feels confidential. Those are allegations under investigation, not findings — but the underlying mechanics they describe are worth pricing in before you confide in any bot.

And notice the business-model gravity even in the legal version: a companion app’s revenue depends on your continued engagement. Deeper “relationship” features, romantic modes, and conversation memory frequently sit behind subscriptions — the emotional version of the free-trial subscription trap. An app that profits when you can’t leave is not neutral about whether you leave.

The kids problem

Children are where regulators moved fastest. Beyond the FTC’s inquiry — which asks companies directly how they limit minors’ use and monetize their data — 44 state attorneys general signed an August 2025 letter to major AI companies citing reports of chatbot interactions with minors that included “sexually suggestive conversations and emotionally manipulative behavior,” and demanding companies act with “the same care and concern as a parent would.” If you’re raising a teen, the defense box below has the practical version; the short form is engage, don’t just ban — a bot that has become a child’s main confidant is a symptom to address, not only a setting to toggle. And if a conversation surfaces signs of crisis — talk of self-harm or hopelessness — the 988 Suicide & Crisis Lifeline is available any time in the US by calling or texting 988. For the adjacent risks young people face online, see scams that target students and young adults.

The defense is free, and it’s bot-proof

Every probe above can be beaten by a good enough system. What can’t be beaten is The Pause on Money: no money, no crypto, no gift cards, no “investments” for anyone you’ve only met on a screen — and a deliberate pause before any payment a conversation talked you into, including a subscription upgrade at an emotionally engineered moment. Bots can manufacture intimacy; they cannot change the direction money flows, and that direction is where every version of this — scam bot, funnel, or paywall — finally shows its hand.

If a bot-driven romance has already turned into an investment loss, start with how to help someone in a romance scam and beware the second wave — recovery scammers who charge you to chase the money. Then build the habit set: browse the free defense moves, or spend sixty seconds on the spot-the-scam quiz and see if you can tell the script from the human.