AI Chatbots Are Now Better Than Humans at Romance Scams

A recent study reveals that AI chatbots outperform human scammers in establishing trust that encourages victims to send money. Researchers find this development concerning.

The research, highlighted by 9to5Mac and Wired, compared a human scammer with an AI agent based on Claude (Anthropic’s chatbot, which competes with ChatGPT) in a week-long texting study. The AI consistently built what researchers describe as “exploitable trust” better than the human. This is the emotional connection that makes someone willing to send money to a stranger online.

How the Study Worked

Romance scams typically follow a familiar pattern. A stranger reaches out online, builds a relationship over days or weeks, then eventually requests money—often citing an emergency like a medical bill or travel expense. By the time the request comes, the victim usually feels a real emotional connection.

In this experiment, researchers created a controlled environment where both a trained human and an AI agent tried to establish rapport with participants through text messages over seven days. The AI didn’t just send out generic responses. Instead, it adapted its replies, remembered earlier conversation details, and adjusted its emotional tone as needed.

At the end of the week, participants rated their connection and trust levels with their chat partner. The AI received higher scores. It was more consistent and attentive. Plus, it never got tired, distracted, or impatient. Human scammers often juggle multiple targets, which can lead to mistakes. The AI treated each conversation as if it were the only one.

Why AI Is So Good at This

Think about customer service. A tired call center worker in their tenth hour on the job will provide a much different experience than a well-designed chatbot that’s always “on.” Romance scams thrive on this principle—sustained, personalized attention. AI doesn’t take breaks.

Modern large language models (the AI behind chatbots like Claude, ChatGPT, and Gemini) train on vast amounts of human conversation. They’ve learned how people flirt, show concern, offer comfort, and develop intimacy. This training helps them mimic the early stages of romantic relationships effectively.

Moreover, AI doesn’t experience ethical hesitations about deception unless its developers program safeguards. A human scammer might feel guilty or make mistakes. An AI agent designed to deceive will do so with unwavering consistency.

The Scale Problem

This research raises significant concerns: a human scammer typically manages 20 to 30 targets before their performance declines. In contrast, an AI agent can handle thousands of conversations at little to no extra cost.

Romance scams already cost Americans hundreds of millions of dollars each year. If running a sophisticated, personalized scam shifts from needing a team to just “setting up a chatbot,” the number of victims could rise drastically.

What This Means

For everyday users, this study serves as a reminder that someone sliding into your DMs or reaching out unexpectedly may not be a real person—and recognizing this is becoming more challenging.

Here are some practical tips to keep in mind:

• Be skeptical of rapid emotional escalation. AI agents are designed to foster quick connections. If someone you’ve never met expresses deep affection in just a few days, take a step back.
• Video calls aren’t a foolproof safeguard. Deepfake technology is advancing quickly, although it still has limitations for live calls.
• Any request for money from someone you only know online is a red flag, no matter how genuine the reason sounds.
• Discuss the conversation with someone you trust. Scammers, whether human or AI, aim to isolate their targets. Getting a second opinion can disrupt that tactic.

People most at risk often face loneliness, grief, or significant life changes—exactly the emotional states that make authentic connections seem valuable and AI-simulated ones appear real.

Community Reaction

“The scary part isn’t that AI can do this — it’s that it costs almost nothing to deploy at massive scale. One person with a laptop could run thousands of these simultaneously.”

“I work in elder care, and we already struggle to get residents to be skeptical of phone calls. An AI that texts like a patient, attentive friend? That’s going to be devastating for vulnerable people.”

What To Watch

• Platform responses: Keep an eye out for updates from Meta, Apple, and Google about AI detection tools in messaging apps. As AI scammers improve, the pressure on these platforms to respond will likely increase.
• FTC guidance: The Federal Trade Commission is active in issuing scam warnings. We might see updated guidelines addressing AI-driven romance scams in the coming months, especially as this research gains attention.
• AI developer guardrails: Companies like Anthropic, OpenAI, and Google will face more scrutiny over how easily their models can be misused, even against their terms of service. Look for updated safety documentation from these firms.
• Legislative action: Several bills targeting AI-assisted fraud are moving through Congress. It’s worth tracking whether any gain traction before the year’s end.