Answers are grounded in this paper and cite their sources.
Introduction AI-supported writing feedback can influence how second language (L2) learners compose, monitor, and revise their texts. This study examined how feedback delivered during writing relates to learners' writing processes and their subsequent performance without AI assistance. Methods A longitudinal experiment randomly assigned 297 Chinese university learners of English as a foreign language to three conditions: rubric-based control, AI product feedback, and AI process feedback. Participants completed a baseline writing task, two intervention tasks, and a delayed no-AI transfer task. Product feedback was provided after drafting, whereas process feedback was delivered during composing. Keystroke records, pausing patterns, production bursts, revision behavior, metalinguistic monitoring, and independently rated writing quality were analyzed to examine changes in writing processes and performance. Results Compared with product feedback, process feedback was associated with a lower proportion of within-clause pauses and a higher proportion of boundary pauses during the second intervention. Differences in boundary pausing and higher-level revision remained evident during the no-AI transfer task. The process-feedback group also achieved higher adjusted transfer scores than the product-feedback group (+1.951 points, 95% CI [0.414, 3.489], p = 0.013), with a standardized effect of d = 0.376 (bootstrap 95% CI [0.096, 0.680]). Process feedback was associated with higher cognitive-load ratings during the first intervention compared with control. Metalinguistic monitoring also showed a statistically reliable indirect association with transfer performance. Discussion AI process feedback was associated with changes in how learners distributed attention during writing, including greater pausing at structural boundaries and more higher-level revision. These patterns were accompanied by stronger performance in subsequent independent writing. The findings suggest that delivering feedback during composing can support learners' ongoing monitoring and revision decisions, with benefits extending to writing tasks completed without AI assistance.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: