Profiles of AI feedback engagement in English-speaking practice: The roles of self-efficacy and growth mindset

AI-generated feedback provides new opportunities for English-speaking practice, but its value depends on how learners process, apply, experience, and manage that feedback. This study examined profiles of AI feedback engagement among 1056 Chinese university EFL students and investigated how English-speaking self-efficacy and growth mindset were associated with profile membership. Latent profile analysis was conducted using cognitive, behavioral, emotional, and ethical feedback engagement as indicators, followed by R3STEP multinomial logistic regression. Four profiles were identified: Minimally engaged, Emotion–behavior active, Moderately balanced, and Reflective-integrated. The Emotion–behavior active profile showed high behavioral and emotional engagement but relatively low ethical engagement, whereas the Reflective-integrated profile combined high engagement across all four dimensions. English-speaking self-efficacy primarily distinguished active engagement from minimal or moderate engagement but did not differentiate the two more strongly engaged profiles. Growth mindset was particularly important for distinguishing the Reflective-integrated profile from the other profiles. These findings demonstrate that active use of AI feedback does not necessarily involve critical evaluation or responsible use. AI-supported speaking instruction should therefore address motivational support, feedback literacy, and ethical AI use together.