Efficiency vs. effectiveness: Self-regulated learning with LLM-mediated help-seeking

Abstract

AI tools powered by large language models (LLMs) are rapidly entering students’ study practices, yet we know little about how they reshape self-regulated learning (SRL), particularly help-seeking and related task strategies. We report a qualitative study with 20 STEM students at a large Swedish university who were interviewed about their use of commercial LLM chatbots. Guided by Zimmerman’s SRL model and the Online SRL subscales, we conducted a thematic analysis. Findings show that LLMs are integrated into a layered, context-dependent help-seeking ecosystem rather than replacing human support. Students described a four-stage process: (1) deciding whether help is needed, typically by attempting problems independently first; (2) choosing a source and using ChatGPT as a low-barrier first step, then peers for conceptual negotiation, and instructors for complex or high-stakes issues; (3) determining the type of help, from seeking hints and explanations to scaffolding problem-solving, streamlining routine work, and extending learning; and (4) judging the help by exercising selective trust, verifying AI outputs against coursework or with humans, and reserving human support for nuanced understanding and affective needs. Overall, students aligned LLM use with SRL goals and task demands, favouring instrumental over executive help-seeking to retain control of problem-solving. The findings suggest an adapted model of help-seeking for LLM-mediated learning practices and implications for promoting verification practices, instrumental help-seeking, and sustained learner agency.