From Automated Correction to Self-Regulation: The Longitudinal Effects of Generative AI-Mediated Feedback on EFL Learners' Revision Behaviors and Writing Self-Efficacy

Generative Artificial Intelligence (GenAI) tools have rapidly become embedded in English as a Foreign Language (EFL) writing instruction, raising urgent questions about whether automated, sentence-level corrective feedback can do more than fix surface errors — specifically, whether it can cultivate the self-regulatory capacities and self-efficacy beliefs that sustain long-term writing development. This narrative review synthesizes evidence from five recent empirical and methodological studies spanning Automated Written Corrective Feedback (AWCF) delivered through tools such as Grammarly, peer-feedback comparisons, Q-methodological investigations of learner cognitive dissonance toward GenAI, and a longitudinal classroom study of real-time affective regulation supported by generative models. Drawing on Social Cognitive Theory and Self-Regulated Learning (SRL) frameworks, the review traces a consistent pattern: automated feedback reliably improves measurable writing accuracy (grammar, lexical resource, coherence, and task response) and produces short-term gains in self-reported self-regulation, but its capacity to generate durable, intrinsically sustained motivation — operationalized in the reviewed literature as Directed Motivational Currents — remains inconsistent and contingent on pedagogical scaffolding. The review further identifies a recurring "reliability-versus-pedagogical-value" tension: learners and instructors regard AWCF tools as technically accurate yet often perceive their feedback as decontextualized, prescriptive, or insufficiently attentive to higher-order rhetorical concerns. Evidence from Q-methodological research on Chinese tertiary EFL learners reveals that this tension manifests psychologically as cognitive dissonance — particularly an efficiency-capacity dissonance in which learners value GenAI's productivity gains while fearing erosion of independent competence — and that learners manage this dissonance through identifiable self-regulation strategies such as sequencing, selective verification, and context-based practice. Synthesizing across study designs, this review argues that revision behavior and self-efficacy gains are not automatic byproducts of automated feedback exposure; they depend on whether feedback is embedded within a structured, hybrid pedagogical model that pairs machine immediacy with human interpretive guidance. The review concludes with an agenda for longitudinal, mixed-methods research examining how GenAI-mediated feedback loops reshape revision behavior and self-efficacy over extended instructional periods, and offers practical implications for designing feedback literacy training in EFL writing classrooms.

کلیدواژه ها:

Generative AI ، EFL Writing ، AI-Mediated Feedback ، Revision Behaviors ، Writing Self-Efficacy ، Self-Regulation ، Longitudinal Effects