Comparing the effects of teacher- and AI-mediated corrective feedback on accuracy, complexity, and quality in L2 written narratives

The rise of generative artificial intelligence (AI) has introduced new possibilities for the implementation of written corrective feedback (WCF), a cornerstone of second language (L2) writing instruction. This study investigates the comparative effects of teacher- and AI-mediated WCF on the complexity, accuracy, and quality of L2 narrative writing. Participants were 166 upper-intermediate ESL undergraduates enrolled in 12 international writing classes at a U.S. university. Of these, 80 received teacher feedback and 86 received AI feedback. All participants completed an initial narrative writing task, revised their texts based on the feedback provided, and submitted final drafts. Analyses revealed that both groups improved significantly in accuracy, with teacher feedback resulting in greater error reduction. Lexical complexity also increased in both groups, with teacher feedback prompting more lexical variation. In contrast, syntactic complexity declined across both groups, marked by shorter T-units and reduced subordination, coordination, and nominalization, but the decline was less pronounced in the AI-feedback group, suggesting a potential advantage of AI in maintaining syntactic complexity. Writing quality improved in both groups, particularly in writing conventions and overall scores, with no significant differences in other dimensions. These findings highlight AI-mediated WCF as a promising complement to teacher feedback, with distinct linguistic outcomes.