<![CDATA[This study investigates whether lightweight AI feedback improves Chinese university students’ oral accuracy in literature-based EFL speaking tasks compared to conventional teacher feedback. A quasi-experimental pretest-posttest design was employed with 64 intermediate-level English majors over 12 weeks. The experimental group received immediate AI-generated feedback on grammatical errors, while the control group received delayed teacher feedback (48 hours). Oral accuracy was measured using error-free clauses (EFC), error-free T-units (EFT), and grammatical accuracy ratio (GAR). The experimental group demonstrated significantly greater improvements across all metrics than the control group, with large effect sizes (d > 1.30). Students receiving AI feedback also reported positive perceptions of its usefulness, ease of use, and intention to continue using such tools. This study uniquely integrates AI feedback with literature-based speaking tasks in Chinese higher education, moving beyond scripted dialogues to examine complex, meaning-oriented oral discourse. AI feedback offers a scalable, immediate, and individualized solution for large EFL classes where teacher feedback is constrained by workload, promoting learner autonomy without disrupting communicative activities. This research provides empirical support for the Noticing and Output Hypotheses, demonstrating that immediate AI feedback accelerates error awareness and grammatical restructuring in oral production, offering actionable insights for EFL curriculum designers and instructors.]]>
