AI-generated feedback in an EAP writing classroom: The collaborative process of feedback, uptake, and revision quality

Abstract

This study examined how EAP learners collaboratively engage with AI-generated feedback on co-constructed texts and to explore how learner engagement with AI-generated feedback influences their uptake behaviors and subsequent individual revisions. Thirty-seven university-level EFL students from two intact EAP writing classes completed two collaborative writing tasks in pairs and received ChatGPT's feedback on their co-constructed texts. They then collaboratively reviewed the feedback and independently produced individual texts on the same prompts.

To examine learner engagement with AI-generated feedback, transcribed peer interactions were analyzed in terms of engagement level. To assess influencing factors, learner perceptions of usefulness of AI feedback and feedback-related factors (i.e., ChatGPT's feedback category and linguistic focus) were also analyzed. Revision quality was evaluated in terms of lexical sophistication, syntactic complexity, and overall accuracy. A generalized linear mixed-effects model and linear mixed-effects models were used to examine uptake behaviors and revision quality, respectively.

Results indicated that learners exhibited varying levels of engagement with AI-generated feedback, from extensive discussion to limited acknowledgement. Higher engagement and more positive perceptions of AI feedback were associated with greater successful uptake. Additionally, greater engagement predicted higher lexical sophistication in revisions, while correct uptake was linked to increased syntactic complexity and greater overall accuracy. Implications of these findings are discussed in terms of implementing AI-generated feedback in EAP instruction.