ABSTRACT Background In Informal digital learning of English (IDLE) environments, recent advances in generative artificial intelligence (AI) have created new opportunities for self‐directed second language (L2) speaking practise through responsive and low‐stakes interaction. However, AI‐mediated IDLE (AI‐IDLE) is not experienced uniformly: learners differ in how they engage with AI, and such differences may be reflected not only in speaking performance and foreign language speaking anxiety (FLSA) but also in the sequential behaviours through which engagement is enacted. Objectives This study aimed to investigate how different profiles of learner engagement in AI‐IDLE are associated with L2 speaking performance, FLSA, and sequential behaviour patterns during AI‐learner interaction. Methods An AI‐IDLE environment was developed to provide scaffolded, avatar‐based speaking practise in contextualised scenarios. Sixty university L2 learners participated in this study. Latent profile analysis (LPA) was conducted based on self‐reported engagement measures to identify distinct engagement profiles. Differences in speaking performance and FLSA were compared across profiles, and sequential behaviour analysis was employed to examine enacted engagement as reflected in AI‐learner interaction sequences. Results and Conclusion Three engagement profiles were identified. Differences across profiles were observed in both speaking performance and FLSA. Sequential analysis further revealed that learners exhibited distinct patterns of interaction with AI, characterised by varying degrees of iteration and continuity in behaviour sequences. These findings suggest that distinct engagement profiles are associated not only with different learning outcomes but also with qualitatively different patterns of interaction with AI. Implications for Practise The study contributes to AI‐IDLE research by integrating person‐centred and process‐oriented perspectives, highlighting the importance of examining both perceived and enacted engagement. The findings offer implications for the design of AI‐IDLE that support sustained interaction, adaptive feedback use, and reduced FLSA.
Chen et al. (Wed,) studied this question.