<p>The integration of generative artificial intelligence (GenAI) into language education presents significant potential, yet empirical evidence on how to incorporate such tools to support learner engagement and enhance outcomes remains relatively limited. Grounded in interaction theory, this study evaluated the effectiveness of a GenAI chatbot-supported flipped classroom model in a college-level English speaking course. A mixed-methods approach was employed over a 15-week intervention, involving 80 Chinese EFL learners assigned to either an experimental group (<i>n</i> = 40), using the GenAI-supported model, or a control group (<i>n</i> = 40), experiencing a traditional flipped classroom. Following an explanatory sequential design, quantitative data (emotional engagement questionnaires, speaking proficiency tests) were first collected and analysed to identify the intervention’s effects. Subsequently, qualitative data from semi-structured interviews with16 students and the course instructor were analysed to explain and elaborate on the quantitative findings. Quantitative results from ANCOVA showed that the experimental group achieved significantly greater improvements in both emotional engagement and speaking proficiency compared to the control group. Qualitative interview data revealed that the GenAI chatbot served as an adaptive scaffold that transformed learner-content interaction by reducing anxiety and building foundational confidence. This affective preparation subsequently enriched the quality of learner-learner and learner-instructor interaction during class sessions, fostering a more supportive and communicatively engaged learning environment. The study concludes that the GenAI-supported flipped classroom offers a theoretically grounded and empirically validated approach for simultaneously developing emotional engagement and linguistic competence in EFL learners, providing practical guidance for educators aiming to integrate GenAI purposefully into language instruction.</p>