AI education study: Behavior trumps background in learning gains

A new study published on arXiv (2607.10101) by Jingwei Yi and nine other researchers from multiple institutions investigates how AI-assisted education benefits different students. They recruited 318 university students for structured learning experiments lasting up to 125 minutes. The key finding: students' learning behavior—specifically proactive and critical engagement versus limited engagement—is strongly associated with learning outcomes. Students who actively questioned, critiqued, and iterated with AI performed significantly better.

Notably, the study found that students from higher-ranking universities and those with greater prior knowledge tended to benefit more from AI assistance. However, when the researchers accounted for learning behavior, these background-related advantages substantially weakened or disappeared entirely. This suggests that the observed benefits are not due to inherent background differences but rather to the fact that these students are more likely to adopt proactive interaction strategies. The findings provide educators with actionable insights: instead of focusing solely on access to AI tools, they should teach all students how to engage critically and proactively with AI to level the playing field.