Students Who Ask Better Questions Learn More From AI, Study Finds
Your kid's chat style with AI predicts their math success—here's how.
In a study of 112 secondary-school students doing open-ended mathematical-modeling practice with a general-purpose LLM, students showed limited observable epistemic proactivity: their interactions were strongly request-dominated, with planning, monitoring, and evaluation relatively rare. Across 880 substantive turns, explicit self-regulated-learning moves increased only modestly, and help-seeking became less understanding-oriented.
Crucially, students whose requests became more understanding-oriented over time performed better on a post-test without AI—beyond pre-test and aggregate patterns. The takeaway: understanding AI-supported learning requires looking at the temporal organization of epistemic proactivity, not isolated prompt types.
- Most students used AI to get quick answers, not to understand concepts.
- Students who asked more 'why' and 'how' questions over time did better on tests without AI.
- The study suggests that how you interact with AI matters more than how much you use it.
Why It Matters
For parents and teachers, it shows that guiding kids to ask deeper questions can boost learning.