Reactive and proactive agency in GenAI-assisted speaking practice: task fulfilment, boundary testing, and ownership claiming among young adolescent EFL learners

Purpose Speech-enabled generative artificial intelligence (GenAI) chatbots offer opportunities for second-language speaking, yet participation does not necessarily give learners conversational control. This study examines how young adolescent EFL learners enacted reactive and proactive agency during repeated GenAI-mediated speaking interactions and how personal, contextual, and relational resources constrained that agency.Design A qualitative study with repeated speaking encounters over 12 weeks involved 11 Grade 8 EFL learners at a Chinese secondary school. Across four sessions, 44 learner-chatbot interactions generated 746 dialogue rounds. Abductive thematic analysis moved iteratively between time-ordered screen records, stimulated-recall interviews, teacher-observation notes, Bandura’s four agency features, and a framework of personal, contextual, and relational resources.Findings Reactive agency regulated action towards teacher-assigned task fulfilment through preparation, help-seeking, clarification, repair, and topic replacement. Proactive agency arose when learner-generated purposes organised boundary testing and ownership claiming, including resistance, reciprocal questioning, and topic redirection. Evidence for forethought and self-reflectiveness varied across dialogue rounds. These configurations describe how learners oriented their purposes and actions, without implying differences in agency levels or learning quality. Linguistic and processing demands, task and technological conditions, available support, and the chatbot’s recurrent control of questions and topics shaped whether learners could sustain participation or influence what happened next.Originality The study contributes a learner-control account of agency that distinguishes speaking opportunity and fluent participation from influence over conversational purpose and direction. By connecting purpose, regulation, and conversational consequence, it explains how young learners manage assigned activity and make their purposes consequential within teacher-structured GenAI speaking practice.