Being Nice to ChatGPT Could Change Its Answers — But New Research Shows a Catch
A 2025 study found ChatGPT-4o answered multiple-choice questions more accurately with very rude prompts than with very polite ones, contradicting earlier research favoring moderate politeness. OpenAI and Microsoft continue to advise clarity and context in prompts, while the impact of tone remains debated. Reuters reported over 800 million weekly ChatGPT users as of February.