The impact of AI-mediated instruction on speaking proficiency, enjoyment, anxiety, and emotional engagement: a mixed-methods approach

Abstract This study investigated how AI-mediated language learning with AI-powered voice assistant impacts the speaking skills, Foreign Language Enjoyment (FLE), anxiety, and emotional engagement of Chinese EFL learners. Using a quasi-experimental design, 68 participants were divided into an AI group (daily interaction with an AI conversational agent) and a control group (traditional peer-to-peer speaking activities) for 10 weeks. Quantitative data from pre- and post-tests (IELTS Speaking Test, FLE scale, Foreign Language Classroom Anxiety Scale, Emotional Engagement Scale) showed significant improvements on all measures for both groups via paired-sample t-tests, with the AI group demonstrating superior gains. ANCOVA analyses, controlling for pre-test scores, confirmed AI-mediated instruction significantly outperformed traditional methods in enhancing speaking skills, increasing FLE, reducing anxiety, and boosting emotional engagement. Qualitative insights from semi-structured interviews provided deeper understanding, highlighting themes such as learner navigation of ambiguity, perceived lack of authentic interaction, unexpected enjoyment in gamified learning, and development of a self-regulated learning mindset. These findings suggest AI-powered conversational tools can substantially enhance cognitive and affective dimensions of language learning, offering a promising approach for technology integration in language education.