From Answer Generator to Questioning Coach: Designing AI-Assisted Coaching for Self-Regulated Learning in Higher Education

Generative artificial intelligence (GenAI) can support self-regulated learning (SRL), but its pedagogical value depends on whether interaction elicits learner cognition or substitutes for it. This study examined how students’ reported use of GenAI corresponded to an intended question-first coaching design, the perceived processes and tensions associated with SRL, and conditions relevant to potential scale. A qualitatively driven embedded case study utilized 31 reflective portfolios as the primary dataset, complemented by 53 survey responses and 68 weekly logs from 12 maximum-variation cases. Abductive, reflexive thematic analysis generated five themes. Question-led interaction appeared most useful when it prompted problem reframing, led to a bounded evidence-producing action, and preserved learner judgment through evaluation, modification, or rejection of AI suggestions. Benefits were qualified by role drift, generic interaction, cognitive substitution, uneven engagement, and continued need for human support. The findings support a distributed coaching-system interpretation in which AI-supported elicitation, pedagogical scaffolding, learner agency, reported AI-role configuration, and function-specific human support interact. The study does not demonstrate institutional scalability; instead, it identifies scale-relevant conditions including sustained engagement, manageable scaffolding, role boundaries, safety, learner judgment, and routes to human expertise.