source-grounded AI

Direct questions to a source-grounded AI helped essay revision, while lasting learning remained untested

Yixin Cheng, Yizhou Fan, Xinyu Li, Guanliang Chen, Dragan Gašević, Zachari Swiecki

Computers and Education: Artificial Intelligence

A study of 59 university students compared essay revision supported by GPT-4 with revision supported by an online human tutor. AI users asked more direct, knowledge-oriented questions and achieved better revised-essay scores. Network analysis linked questioning patterns to that difference, whereas simpler frequency-based models did not show the same mediation. The study measured performance while assistance was available, leaving independent learning and lasting transfer unresolved.