Students who work through problems with a chatbot know more facts to say, but say less and share fewer opinions than with classmates.

This explores what changes when a student works through a problem with a chatbot instead of a classmate, both in how much they say and in what kind of thing they say. The corpus has one direct head-to-head study, and it shows a trade. Students working with chatbots did better on practical tasks and produced more knowledge-based dialogue than peer groups did. But they talked significantly less overall and voiced far fewer subjective perspectives, meaning personal takes, reactions and opinions Does chatbot interaction trade authenticity for better problem-solving?. A chatbot gets students to talk about the material, but they talk less as themselves.

This cuts against what the corpus says about chatbots elsewhere. People disclose more intimate material to a chatbot because no one is judging them Do chatbots help people disclose more intimate secrets?. Machines also strip out secondary social goals like face-saving and impression management, and replace them with simpler goals like being understandable Why do people share more openly with machines than humans?. You might expect students to open up more with a chatbot, yet they said less of themselves. One way to reconcile this is my inference, and the corpus doesn't test it. The lack of social stakes that makes a confessional easy also removes the reason to voice an opinion. A classmate is an audience you want to persuade, impress or bond with, while a chatbot is a resource you extract answers from.

How chatbots sound may reinforce this. Their language carries signals of expertise that move users from actively searching and recalling to relying on the system, and trust attaches to the confident register rather than to accuracy Does chatbot language style actually shape how much we trust it?. A student who treats the chatbot as the expert will ask, refine and apply, but has little reason to argue or riff. The exchange is also lopsided. AI output only looks like an utterance because the human supplies the missing orientation through interpretive labor, so the structure of the exchange exists only on the student's side Does AI generate genuine utterances or just text patterns?. AI can predict social norms better than any individual person, but it can't take part in the community process where norms and shared knowledge get made Can AI predict social norms better than humans?. A human peer is someone with a stake in the answer who pushes back.

This isn't fixed by the medium, though. Users reciprocate when a chatbot shares emotions consistently, so a chatbot's own behavior can draw out more personal expression Do chatbots trigger human reciprocity norms around self-disclosure?. In a simulated classroom, different counselor-chatbot styles sent student stress and AI dependence down different paths over weeks. The effects came through the actual replies rather than the style label, and they spread through peer interactions How do different counselor styles shape student stress and AI dependence?. So chatbots reshape how students talk to each other, not only how they talk to the bot. There is also a design route that keeps the AI in a peer-like role: AI teammates can hold natural, human-like conversation while steering toward observable evidence of collaboration skills Can AI teammates assess collaboration without losing naturalness?.

Treat the trade-off as provisional. Novelty effects in chatbot relationships fade predictably, and single-session findings don't reliably extend to longer use Do chatbot relationships lose their appeal as novelty wears off?. The student study may therefore be measuring a first-encounter pattern. Whether students would eventually bring their own voice to chatbots, or stop doing so, is something this corpus can't yet answer.