Reframing GenAI-mediated EFL speaking: A systematic review of empirical studies through an Activity Theory lens

The rapid development of generative artificial intelligence (GenAI) has created more possibilities for English as a foreign language education. Through its generative capabilities, particularly in natural language generation and processing, GenAI has introduced new modes of interaction for learners’ speaking development. However, empirical research on GenAI-mediated EFL speaking remains fragmented. Guided by PRISMA 2020, this systematic review synthesizes 43 empirical studies published between 2024 and 2026 to examine the trends, application methods, and challenges of GenAI-mediated EFL speaking activities. Findings indicate that research has focused primarily on mainland China and intermediate-level undergraduate learners, with ChatGPT as the most frequently used tool. GenAI was commonly used to provide personalized practice and immediate feedback in a comparatively low-pressure environment. Most studies reported positive outcomes, but findings varied. Behavioral outcomes received less attention than cognitive and affective aspects. Challenges mainly involved inauthentic interaction and limited feedback. Tensions are mainly from tools and subjects, as well as tools and objects from an Activity Theory lens. Based on these tensions, this study proposes a collaborative learning framework to provide complementary mediation in EFL speaking activities.