How Do Students Regulate Their Learning With a GenAI Chatbot?

How do students regulate their learning with a genAI chatbot? — an exploratory trace-based study of 30 university students who read texts and wrote an essay with an integrated ChatGPT chatbot. Most students (73%) voluntarily used the chatbot, mainly for seeking information, and chatbot users achieved higher essay scores than non-users — but qualitative evidence showed students offloading comprehension and synthesis to the chatbot, outsourcing key regulatory decisions.

Key Findings

• Chatbot as information source and surface-task helper. Queries were mostly seeking information (53%), then language support (25%), generating ideas (13%), and correction/repetition (9%). The chatbot was used ~12% of learning time.
• Performance benefits but outsourcing risks. Chatbot users scored higher on the essay, and chatbot interaction frequency correlated positively with high-cognitive writing activities. But students asked the chatbot to "summarize the paragraph into 3 sentences" or "extract only the main ideas" — offloading comprehension, synthesis, and key regulatory decisions (evaluating text relevance, deciding what to read, integrating new knowledge).
• Low planning and monitoring. Both chatbot users and non-users engaged in almost no planning and monitoring; the chatbot potentially replaced reading, a core activity for building an initial knowledge base.
• Production deficit. Students possessed sufficient SRL knowledge and skills but failed to spontaneously engage in AI Regulation in Education during learning.

What this means for practice

• Instructional designers. Build monitoring prompts into the tool: when a query signals offloading — "summarize the paragraph into 3 sentences" or "extract only the main ideas" — require the student to reflect on their own reading and comprehension first.
• Instructional designers. Configure the assistant for educational fit rather than task efficiency, as this study did with constraints such as not writing the essay, keeping responses succinct, and grounding suggestions in the provided texts.
• Instructors. Do not treat higher essay scores as evidence of learning: chatbot users outscored non-users on the essay while their queries showed comprehension and synthesis being outsourced, and both groups did almost no planning or monitoring.
• Educators. Make reading visible and assessed, because longer chatbot use and high-cognition interaction correlated negatively with reading time — the activity that builds the knowledge base students then synthesize.

Limitations

• Thirty students from a single university seminar on social science research methods (M age = 25.70, SD = 2.88) completed one 45-minute reading-and-essay task, so the findings are exploratory and tied to that setting.
• Chatbot use was voluntary rather than assigned, making the comparison between users (73%) and non-users self-selected rather than randomized.
• SRL was inferred from trace-based event analysis and coded chatbot query logs, not from direct measurement of students' planning, monitoring, or strategy use.
• All participants reported prior experience with genAI tools, so the sample does not represent novice users, who may offload differently or less skillfully.

Citation

Lim, L., & Bannert, M. (2026). How Do Students Regulate Their Learning With a GenAI Chatbot?. Learning Letters, 8, Article 61.