Beyond fluency

A systematic review of AI-mediated language learning and intercultural competence in higher education

Abstract

AI-mediated language learning is increasingly used in higher education, yet its contribution to intercultural competence remains insufficiently synthesised. This systematic review examines how AI-mediated technologies and pedagogical designs have been used, which intercultural outcomes have been reported and assessed, and what conditions shape their contribution. Following PRISMA 2020, Scopus, Web of Science Core Collection, and ERIC were searched for peer-reviewed English-language journal articles published from January 2023 to July 2026. From 180 records, 25 primary empirical studies met the eligibility criteria and were examined through narrative thematic synthesis. ChatGPT, large language models, and conversational agents were the dominant technologies, but technology type alone did not determine outcomes. Clearer intercultural development was reported when AI was embedded in culturally contextualised simulations, reflective or metacognitive scaffolding, collaborative exchange, and authentic human interaction. Reported outcomes covered cognitive, affective, and behavioural dimensions, although evidence was more consistent for cultural awareness, perspective recognition, openness, confidence, and willingness to communicate than for sustained pragmatic or behavioural transfer. Assessment approaches varied considerably and frequently relied on self-reported measures. AI offered accessible practice, immediate feedback, and cultural inquiry, but cultural simplification, bias, overreliance, unequal access, and limited interactional repair constrained its value. AI is therefore best understood as an intercultural learning scaffold rather than a cultural authority or substitute for human interaction. Explicit intercultural objectives, critical verification, human mediation, and performance-based assessment are needed to move language learning beyond fluency.