A Structural Model of Artificial Intelligence Integration in Language Instruction: A Systematic Review of Mediating Pathways to Written English Proficiency Among EFL Learners

Abstract

Background: Generative artificial intelligence (AI) is reshaping English-as-a-foreign-language (EFL) writing instruction, yet the mechanisms through which AI integration may improve written English proficiency (WEP) remain theoretically fragmented. Objective: This systematic review synthesises literature published within the 2018–2026 review window, together with foundational theoretical works, to develop a structural-equation-modelling (SEM) framework in which self-regulated learning (SRL), learner engagement, self-efficacy, and AI literacy operate as parallel and serial mediators between AI integration and WEP. Method: Following a structured thematic-synthesis approach, the reviewed literature was organised around four connected domains: AI-supported writing instruction, generative-AI adoption, learner and teacher perceptions, and self-regulated learning in AI-supported contexts. The synthesis then examined evidence for direct, mediating, and serial relationships among the focal constructs. Key findings: The literature provides partial support for the individual links connecting AI integration, the proposed mediators, and language-learning outcomes; however, these pathways are rarely tested within a single integrated model. Evidence for mediation remains fragmented, the proposed serial chain is largely untested, and Jordanian and broader Levantine contexts remain under-represented relative to East Asian and Gulf settings. Implications: The review proposes fourteen hypotheses: nine direct-effect hypotheses (H1–H9) and five mediation hypotheses (H10a–H10e), including one serial mediation chain, as a framework for future quantitative validation among English majors at public Jordanian universities.