COMPARATIVE EFFECTIVENESS OF AI-GENERATED AND PEER FEEDBACK ON L2 WRITING PERFORMANCE: IMPLICATIONS FOR PAKISTANI EFL CONTEXTS

Abstract

This study compared the effectiveness of ChatGPT-generated feedback and structured peer feedback in improving Pakistani undergraduate EFL learners’ L2 writing and examined how learners’ perceptions and feedback uptake help explain writing outcomes. The study used a mixed-methods sequential explanatory, quasi-experimental design with two intact groups (AI feedback vs rubric-guided peer feedback). A total of 72 university EFL students participated (36 per group). Over an 8-week draft feedback revision cycle, learners completed a pre-test and post-test writing task scored with an analytic rubric (Task Response, Coherence/Cohesion, Lexical Resource, Grammar Accuracy). Feedback artifacts were retained and coded for uptake (accepted/modified/ignored) and revision focus, alongside a post-intervention perception survey and follow-up interviews. Quantitative analyses (ANCOVA controlling for pre-test performance) indicated that both groups improved, and overall post-test writing performance was broadly comparable between conditions. However, improvements differed by dimension: the peer-feedback group showed stronger gains in Task Response and Coherence/Cohesion, while the ChatGPT-feedback group showed stronger gains in Grammar Accuracy (with a smaller advantage in lexical improvement). The AI group produced and implemented more feedback items (including more accepted changes) but also ignored more feedback; uptake rates were similar. Learners rated AI feedback as clearer, whereas peer feedback yielded higher overall satisfaction. Findings support a hybrid feedback model in Pakistani EFL writing classes, combining peer feedback for higher-order discourse development with AI feedback for language accuracy, alongside explicit instruction in feedback literacy and critical evaluation of AI suggestions.