The Teacher-Guided Generative AI for Sustainable EFL Writing Development

Longitudinal Validation of the AI-MWD Framework

Abstract

The rapid adoption of generative artificial intelligence (GenAI) has transformed English as a Foreign Language (EFL) writing instruction by providing immediate, individualized, and context-sensitive feedback. Despite these advantages, increasing evidence suggests that unguided interaction with GenAI often overwhelms learners' cognitive capacity, encourages passive dependence on AI-generated revisions, and weakens the development of long-term writing autonomy. This study addresses this challenge by proposing and empirically validating the AI-Based Metacognitive Writing Development (AI-MWD) Framework, a teacher-guided instructional framework that integrates macro-pedagogical scaffolding with AI-enabled micro-scaffolding to promote sustainable writing development. A 14-week longitudinal quasi-experimental study was conducted with forty undergraduate EFL learners assigned to either an experimental group receiving the AI-MWD intervention or a control group using the same GenAI tools without structured teacher guidance. Independent-samples t-tests examined differences in cognitive load and metacognitive writing autonomy, while 2 × 3 mixed-design ANOVAs evaluated longitudinal changes in Complexity, Accuracy, and Fluency (CAF) across pre-test, post-test, and delayed post-test measures. The findings revealed that teacher-guided AI significantly reduced learners' extraneous cognitive load while substantially enhancing metacognitive writing autonomy. Although both groups demonstrated comparable improvements in writing accuracy immediately after the intervention, only the AI-MWD group maintained these gains six weeks later, indicating significantly stronger long-term retention of linguistic knowledge. Unlike previous AI-assisted writing studies that primarily report immediate performance improvements, this study demonstrates that teacher-guided scaffold fading is the critical mechanism through which AI-supported feedback becomes internalized as learners' metacognitive writing strategies. The proposed AI-MWD Framework therefore reconceptualizes generative AI not as an autonomous instructional agent, but as a pedagogically mediated cognitive scaffold capable of supporting sustainable second-language writing development.

Keywords: Generative Artificial Intelligence, Teacher-Guided Instruction, Metacognitive Writing, Cognitive Load, Scaffolding, Writing Autonomy, Longitudinal Learning, EFL Writing