Generative AI feedback in EFL writing instruction: a mixed-methods study on performance, motivation, and self-regulated learning

With the rapid development of generative artificial intelligence, AI feedback in EFL writing instruction has gained increasing attention. Grounded in social cognitive theory and self-regulated learning theory, this mixed-methods study compared AI-mediated and teacher-mediated feedback delivery modes and examined their relationships with writing performance, learning motivation, and self-regulated learning among 120 s-year English majors at Henan Normal University. A 16-week quasi-experimental pretest-posttest-delayed test design was combined with semi-structured interviews. Repeated measures ANOVA and mediation analysis (PROCESS macro) were used for quantitative analysis; thematic analysis was applied qualitatively. The AI-mediated feedback condition was associated with greater improvement in writing performance (group × time interaction: p < 0.001, η2p = 0.12), with the largest effect on grammatical accuracy (η2p = 0.16), higher writing self-efficacy (4.42→5.82), and reduced writing anxiety (3.95→2.78). Learning motivation and self-regulated learning positively correlated (r = 0.64, p < 0.001) and partially mediated the feedback condition to performance relationship (indirect effects: b = 0.38 and b = 0.45, respectively). Qualitative themes revealed learners’ adaptive progression from passive acceptance to critical AI usage. These findings extend feedback theory to the digital context and offer an integrated framework for AI-assisted language learning.