Student-Reported Cognitive Engagement with Structured Generative-AI Questioning: Differentiated Associations with AI, Teacher, and Peer Questions

Abstract

Generative AI can sustain follow-up questioning, but questions do not constitute engagement by themselves. This qualitative-dominant mixed-methods study examined student-reported cognitive engagement around AI, teacher, and peer questions in a Chinese university critical-thinking course. Seven reflection points produced 1034 matched submissions, including 648 substantive texts. A structured source item provided an implementation check for 399 responses at two common reflection points, and two human coders independently analyzed a stratified sample of 120 reflections. Sampling-weighted and reflection-stratified checks assessed sensitivity to unequal stratum sizes. Within this structured intervention, AI-salient accounts more often concerned assumptions, criteria, evidence, uncertainty, and reason organization; teacher-salient accounts made situated clarification more visible; and the smaller peer-salient set highlighted comparison and counter-positioning. These patterns changed little after weighting, although the AI prompt explicitly requested several of the functions later observed in AI-salient reflections. Across sources, reason organization was clearer than low-pressure participation. The reflections did not establish inherent source effects, complete interaction sequences, critical-thinking gains, or a consistently stronger hybrid pattern. They support a testable proposition: question sources may contribute complementary reasoning resources when students remain responsible for evaluating and revising their judgments.