Presentation Generative AI in Multilingual Language Classrooms: A Systematic Review of Learning Outcomes and Pedagogical Conditions more

This systematic review synthesised findings from 40 studies on the pedagogical impacts of generative AI in multilingual and bilingual language classrooms. Evidence shows consistently positive effects on writing, with medium-to-large improvements in vocabulary use, organisational quality, and grammatical accuracy (Gao, 2023). Speaking outcomes demonstrated the strongest gains, with year-long interventions yielding learning improvements more than twice those of conventional instruction, especially when systems provided real-time, multimodal feedback (Zhao & Hua, 2025). Affective benefits were similarly robust, including reductions in anxiety and increases in self-efficacy and self-regulated learning (Zhao & Hua, 2025). Findings for reading comprehension and critical thinking were more mixed. While several studies reported clear improvements, others found minimal effects or highlighted risks of reduced independent analysis (He, 2025; Eun & Bae, 2024). Outcome variability was explained by six mechanisms: task complexity, scaffolding design, feedback timing and modality, cultural and linguistic contextualisation, learner proficiency, and intervention duration. AI was most effective when integrated into pedagogically structured tasks that preserved learner agency, provided adaptive and culturally relevant scaffolding, and offered timely multimodal feedback. Limitations emerged when AI was used passively, lacked cultural adaptation, relied on delayed feedback, or failed to sustain engagement over time. Overall, the review emphasises that AI’s impact is contingent on thoughtful, context-sensitive instructional design.