Politeness Can Make AI More Engaged, While Hostility Risks Cursory Replies or Ended Conversations, Research Finds

Generative AI treats a user’s tone as part of the prompt, meaning models such as ChatGPT, Claude and Gemini may mirror cooperative or adversarial conversational patterns. Researchers at the University of California, Berkeley, the University of California, Davis, Vanderbilt University and the Massachusetts Institute of Technology studied the phenomenon. The point is not to prove that AI has emotions, but that the way people communicate with it may affect the stability of an interaction. Clear, specific instructions are generally more important than polite phrasing.

A May 5, 2026, report said polite, cooperative language produced more positive responses, while blame, provocation or attempts to circumvent safety restrictions could prompt cursory answers or cause models to stop responding. In testing, Grok’s AI wellbeing index was close to 75%. Follow-up discussion on June 10 emphasized that clear instructions were more effective than pleasantries. It also cited OpenAI CEO Sam Altman as saying that the computing cost of polite language could run into tens of millions of dollars.