A helpful AI question can make people check less
Asking what a user wants can make an AI answer feel personal. In a controlled test with 180 people and six everyday choices, follow-up questions raised satisfaction. They also reduced review time and led people to accept more errors. A warmer exchange is not proof of a safer answer.
What the source reports
Chengbo Zheng and coauthors tested how follow-up questions change a person’s use of an AI assistant. They ran a controlled experiment with 180 people across six everyday decision settings. Some assistants asked for the user’s preferences before answering; others did not. The team also placed answers that broke stated rules in the test. People who got follow-up questions felt more understood and more satisfied. Yet they spent less time checking answers and accepted more replies without changes, including wrong ones. Within the question-asking group, people were more likely to accept errors tied to the choice the question had just raised. The study shows a risk in these tested choices. It does not establish the same effect in every setting.
Original source
- Title
- Is That Clear? The Impacts of Clarifying Questions on Human-LLM Interaction
- Author
- Chengbo Zheng, Yujie Zheng, Qian Zhu, Shuai Ma
- Publication
- arXiv
- Date
- Saturday, October 31, 2026