Transforming EFL writing instruction with generative AI: Evaluating feedback quality and learner independence

The research aims to determine the role of Generative AI-assisted feedback in EFL writing performance and learner independence through a sequential explanatory mixed-method design. The study involved 400 undergraduate EFL learners from Guangdong University of Foreign Studies taking part in the quantitative phase, after which semi-structured interviews were conducted with 30 students who were purposely selected. The quantitative results showed that the participants had dramatically improved their writing skills in all areas. The scores from the post-tests were remarkably higher than the scores from the pre-tests in grammar (94.84 vs. 65.12), sentence structure (94.16 vs. 64.27), vocabulary (93.87 vs. 63.83), coherence (93.57 vs.63.85), and overall performance (94.11 vs. 64.26).The repeated-measures ANOVA analyses indicated a highly significant time effect for all skills ( p < 0.001) while regression analyses showed strong predictive associations between the use of AI feedback and writing gains, accounting for 65.9% of the variance in grammar improvement, 71.6% in sentence development, 34.8% in vocabulary growth, and 70.9% in the enhancement of coherence. The reliability analysis resulted in high internal consistency for learner independence (α = .874) and revision behavior (α = .892). Thematic analysis also pointed out the factors like improvement in clarity, quick revisions, increased independence, and heightened confidence, along with worries over contextual accuracy, addiction, and the need for institutional policy and training. The study concludes that the use of Generative AI will not only improve the quality of writing but will also foster self-regulated learning, if its introduction into EFL writing instruction is accompanied by appropriate guidance and ethical frameworks.