Answers are grounded in this paper and cite their sources.
AI systems can respond to nearly every learner turn, but how learners use that feedback during an exchange is less well documented. This study followed 48 university EFL learners across eight text-based classroom sessions using GPT-4o (384 learner-session observations), examining participation, words per turn (WPT), and researcher-coded next-turn uptake. Following each session, learners submitted ChatGPT-generated summaries of their messages, while researchers separately coded immediate revisions from the interaction records. Mean learner turns increased from 12.3 to 14.5, WPT from 29.3 to 34.6, and mean learner-session uptake from 0.36 to 0.68. Mixed-effects models showed positive session associations with WPT (B = 0.663, p < .001) and uptake (logit B = 0.181, p < .001). After accounting for the session trend, a supplementary model did not identify a distinct phase-level uptake difference following the midpoint prompt revision (B = 0.153, p = .327). The study documents increasing immediate use of AI feedback during repeated interaction, although the fixed phase order prevents separating prompt refinement from accumulated classroom experience.
No takes yet. Share an insight, caveat, or question.
Jin Tang (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: