Designing GenAI-supported input–output integration for EAP learning

The rapid development of GenAI is reshaping EAP instruction as preparation for EMI learning challenges. This study investigates an AI-supported instructional design with iterative revision conditions to scaffold the integration of academic listening and speaking development. Grounded in Skill Acquisition Theory, the study conceptualizes learning through an input-output integrated model emphasizing the transition from declarative to procedural revision. A mixed-methods pretest-posttest design was adopted with triangulated quantitative and qualitative data sources. Participants were university freshmen enrolled in a required EAP training program integrating video-based instruction, speaking practice, and adaptive AI feedback across iterative revision cycles. Quantitative data included proficiency scores and survey responses analyzed through t -tests and cluster analyses, while qualitative data included interaction logs and reflective responses analyzed through content and thematic analyses. Results showed substantial proficiency improvement in both revision conditions with large effect sizes, although no significant between-group differences were observed. Lower-proficiency learners demonstrated the largest gains over time, suggesting that AI-supported instruction may be particularly beneficial for students with weaker initial proficiency. Content analysis revealed that output refinement was the dominant interaction pattern, followed by AI support and error awareness, suggesting a gradual shift toward performance-oriented AI collaboration. Learners also perceived AI as a supportive partner for fluency, vocabulary development, pronunciation, and spoken language organization. The study contributes to understanding how GenAI can support input-output integrated EAP instruction and AI-mediated academic language development.