Generative AI in Efl Speaking and Pronunciation Practice: A Systematic Review of Challenges and Mitigations

While the role of generative artificial intelligence (GenAI) in English as a foreign language (EFL) writing instruction has received growing research attention, its impact on speaking and pronunciation remains underexplored. This PRISMA-based systematic review examines the challenges associated with using GenAI tools, particularly large language model (LLM)-based chatbots, for EFL speaking and pronunciation practice and synthesizes evidence-based mitigation strategies. Twelve studies that met the eligibility criteria yielded 48 challenges. These were organized into six thematic categories, including technical, pedagogical, learner-related, and equity concerns. The studies also varied in research design and instructional context. Several recurrent patterns emerged across the categories. Output accuracy and reliability represented the most persistent challenge and were linked to concerns about learner trust and uncritical reliance. AI-generated feedback was also repeatedly described as mechanical, formulaic, and lacking contextual sensitivity, highlighting a tension between the accessibility and scalability of these tools and limitations in areas that are important to learners. The studies commonly identified a human-in-the-loop approach as a key mitigation principle, alongside technical improvements and pedagogical interventions, including prompt literacy training. However, important questions remain unresolved, particularly whether GenAI reduces or increases teacher workload and if these tools narrow or widen existing equity gaps. Overall, the findings suggest that effective integration of GenAI into EFL speaking instruction requires attention to technical design, pedagogy, and institutional support, while several aspects of its implementation remain insufficiently understood.