From tools to pedagogical ecologies: a systematic review of artificial intelligence in EFL/ESL education

Abstract

Artificial intelligence (AI) is increasingly embedded in English as a Foreign/Second Language (EFL/ESL) education, yet existing research remains fragmented across tools, skills, and short-term outcomes. This review examines how AI is integrated into EFL/ESL education across language skills, instructional domains, and educational contexts. This PRISMA-guided systematic review synthesised 221 unique peer-reviewed publications. Moving beyond a tool-centred inventory, the review analysed AI through four interrelated dimensions: pedagogical roles, mediating processes, reported outcomes, and contextual constraints. AI systems increasingly operated as multifunctional pedagogical actors rather than isolated instructional aids. The most frequently coded roles were teacher orchestration/support, content and materials generation, assessment or diagnosis, coaching or practice companionship, tutoring or scaffolding, and conversational partnership. AI-mediated learning was especially concentrated in writing and speaking/communication, where text-based, voice-based, multimodal, immersive, and adaptive systems supported feedback, revision, rehearsal, and learner–system interaction. Reported benefits included expanded practice opportunities, accelerated feedback cycles, redistributed instructional labour, skill development, learner autonomy, affective support, assessment and monitoring, collaboration, and multilingual engagement. Recurring challenges included technical reliability and feedback quality, teacher readiness, privacy and data security, learner over-reliance, infrastructural inequality, bias and cultural mismatch, authorship and academic-integrity concerns, and methodological weaknesses. Interpreted through a three-layer framework of efficiency, pedagogy, and ideology, the synthesis conceptualises AI in EFL/ESL education as a pedagogical ecology in which tools, learners, teachers, feedback regimes, assessment practices, institutional infrastructures, and governance arrangements interact. The review provides a theory-informed framework for analysing, designing, and governing AI-mediated language education beyond simple claims of technological effectiveness.