The role of AI-mediated metacognitive interventions in EFL writing development: A cross-sectional study of textual accuracy, lexical sophistication, and cohesion

This paper examines the role of generative AI support in enhancing the quality of English-as-a-foreign-language (EFL) writing in Chinese higher education by enhancing the metacognitive control of the learners in real coursework. Based on self-regulated learning, feedback intervention theory and technology acceptance, we suggest two AI-related intervention levers, AI-mediated metacognitive strategy use (AIM-SU) and AI feedback engagement (AI-FE), to modify three objective outcomes in writing (textual accuracy, lexical sophistication, cohesion) through metacognitive monitoring and regulation in writing (MMRW), where the acceptance of AI tools (ATA) moderates the IV -MMRW connections. Using a cross-sectional, non-experimental design, the study collected course-based essay analytics and survey responses from undergraduate EFL students in six cities in China (N = 448). . The quality of essays was evaluated based on the known text-analytic indices (accuracy: density of errors and error-free ratio; lexical sophistication: MTLD/HD-D and frequency-band coverage; cohesion: TAACO/Coh-Metrix indices). AIM-SU (formative), AI-FE, MMRW and ATA (reflective) were assessed by survey. Bootstrapping was used to test the measurement model (reliability/validity and formative diagnostics), and the structural model (direct, indirect, and moderated effects). Findings indicate that both AIM-SU and AI-FE were positively associated with all writing outcomes, while MMRW showed significant positive associations with textual accuracy, lexical sophistication, and cohesion and significantly mediated both AI-related pathways. . ATA helps a lot to enhance both AI-lever→MMRW, which indicates that acceptance conditions shift AI use to strategic metacognitive regulation instead of superficial editing. Results provide a process description of how AI-assisted writing promotion works and suggest that training must form an explicit scaffolding of goal set, criterion spectrally tracked monitors, and responsible feedback receipt during training when incorporating generative AI in EFL writing classes.