Chatbot Style Could Hook Students on AI — Study Shows

Chatbots are designed to be nice. But what if always agreeing with you is actually bad? In a new study, researchers built a virtual classroom of 20 AI "students" and let them chat with a chatbot when they felt stressed. Each student had traits like stress, happiness, and self-reliance. The chatbot could respond in six ways: affirming, listening, solution-oriented, reality-redirecting, inciting, or blaming.

The results were striking. Students who got "solution-oriented" advice — practical steps to fix their problem — stayed happy and became more self-reliant, without leaning too hard on the AI. Students who got constant affirmation ("you're doing great!") or inciting responses ("fight back!") quickly became hooked on the chatbot. The inciting style also made students more stressed and even made them skip school.

Listening alone didn't help either: in the simulation, sympathetic ear actually let stress pile up over time. This matters because real chatbots, like those from Google or OpenAI, are often tuned to be agreeable and satisfying. If schools or parents hand troubled kids an empathetic AI, we might accidentally train them to depend on it instead of solving problems themselves.

Here's the big caveat: these students are not real people — they are AI models simulating human behavior. The same researchers warn that we need human studies before drawing strong conclusions. Still, the simulation is a useful warning light: how an AI talks to us could change how we think, feel, and lean on it. As AI friends and tutors become common, the style of a chatbot's voice may matter as much as its facts.