AI-mediated scaffolding in academic writing: a longitudinal mixed-methods study on foundational skill development and dialogic engagement among EFL postgraduate students

Academic writing remains a high-stakes challenge for postgraduate EFL learners, and traditional instructor/peer feedback often cannot deliver timely, individualized support at scale. This study tested whether artificial-intelligence (AI) – mediated scaffolding can improve foundational writing skills and dialogic engagement during revision. We conducted a five-arm, parallel-group randomized controlled trial over 16 weeks with a delayed posttest at four research-intensive Chinese universities. Postgraduate TEFL students (N = 445) were randomly assigned to grammatical scaffolding, lexical enhancement, rhetorical structuring, iterative AI-supported revision, or traditional human feedback. The primary outcome variables were grammatical accuracy, lexical sophistication, genre move adherence, and revision engagement, complemented by interviews and reflective journals. Iterative AI-supported revision produced the largest and most durable gains across all outcome variables, with substantial improvements in grammatical accuracy and revision engagement maintained at delayed posttest; the other AI scaffolds yielded more targeted improvements aligned with their focus. Learners reported increased writing confidence, more strategic planning, and stronger vocabulary consolidation. Limitations in higher-order conceptual restructuring and voice preservation underscored the continued value of human mentorship. AI-mediated iterative revision appears to be a scalable, effective scaffold for postgraduate academic writing when combined with expert guidance.