Teachers’ AI-TPACK, pedagogical judgment, and students’ learning outcomes in GenAI-mediated L2 writing: a multilevel mediation study

Purpose With the rapid and widespread adoption of generative artificial intelligence (GenAI) in modern L2 writing instruction, teachers face an urgent need to convert GenAI-oriented capacity into ethical and process-based instructional support.Design Based on the writing-oriented Artificial Intelligence–Technological Pedagogical Content Knowledge (AI-TPACK) framework and the boundary-making perspective, this study constructs a two-wave (T1–T2) multilevel mediation model in which teachers’ AI-TPACK and scenario-based pedagogical judgment (SPJ) indirectly affect students’ learning outcomes through classroom-shared perceived instructional quality (PIQ), operationalised as the between-class component of student-reported instructional quality. The sample included 51 university EFL writing teachers and 513 students from five research-intensive universities in Eastern China, complemented by interviews with 12 teachers.Findings Change-adjusted analyses yielded the following findings. First, both professional capacities were positively correlated with PIQ. Second, PIQ was positively associated with students’ writing ability, indexed by self-efficacy and self-regulation. This pattern was also supported by blind ratings of T2 writing tasks completed without GenAI assistance. Interview data indicated that effective GenAI-mediated instruction relied on routinised boundary clarification, explicit criteria for acceptable GenAI use, and verification-oriented accountability.Originality/Value This study highlights that teachers’ professional capacities matter for L2 writing development through PIQ, understood as students’ shared perceptions of enacted instructional practice.