Inquiry-Based Learning Patterns in Large Language Model-Driven Learning Environments: An Exploratory Study From Bloom's Perspective

Synthesis: This exploratory study of students interacting with LLMs during an inquiry-based learning data-science writing task used Bloom's taxonomy to analyze behavioral patterns across cognitive levels. Thematic analysis of 117 interview transcripts, 370 interaction records, and 1,694 minutes of screen recordings identified 14 interaction patterns at different levels of prior knowledge. The study highlights how Self-Efficacy and metacognitive monitoring shape LLM use in IBL and calls for guiding planning frameworks and Scaffolding to mitigate AI reliance.

Key Findings

1. 14 interaction patterns across Bloom's cognitive levels. Students interacted with LLMs in 14 distinct patterns, varying by level of prior knowledge.
2. Prior knowledge shapes interaction. Students at different prior-knowledge levels used LLMs differently during the IBL process, affecting how they progressed through cognitive stages.
3. Self-efficacy and metacognitive monitoring matter. These factors significantly influenced learning behavior in the LLM-driven environment — and their absence correlates with reliance on AI.
4. Design implication: scaffolding for higher-order thinking. Educators should design cognitive scaffolding targeting specific higher-order thinking stages; instructional designers should build planning frameworks that mitigate over-reliance while fostering metacognitive monitoring.
5. Policy: critical evaluation training. Policymakers should implement training to enhance students' critical evaluation skills within LLM-driven environments.

What this means for practice

• Instructors. Aim Scaffolding at specific higher-order thinking stages rather than at LLM use in general, because prior knowledge and metacognitive skill determine whether LLM use deepens inquiry or bypasses it, and the 14 patterns vary by prior-knowledge level.
• Instructors. Teach critical evaluation of LLM output inside the inquiry task itself, since the study's policy implication is explicit training in evaluating responses within LLM-driven environments.
• Designers. Build planning frameworks that keep metacognitive monitoring active and mitigate over-reliance, because Self-Efficacy and metacognitive monitoring shaped learning behavior in the LLM-driven environment and their absence correlated with reliance on the AI.
• Researchers. Follow the exploratory design with larger, objective measurement — the authors plan quantitative work with a larger sample and eye-movement data — because the current design cannot clarify causal links between behaviors such as cue design and critical thinking.

Limitations

• Only 19 students in a single data-science IBL course at one Chinese university took part, so the 14 interaction patterns describe one restricted sample rather than a general population.
• The study cannot clarify the causal relationship between specific behaviors, such as cue design and critical thinking, and the authors state it may lack an assessment of the unique impact of the LLM environment.
• It recorded one IBL course only, so it may not capture long-term behavioral change or how cognitive levels shift in sustained LLM-assisted learning.
• Analysis rested on prompt text and retrospective think-aloud interview transcripts, with screen recordings used to verify rather than as a primary measure; retrospective accounts can miss or rationalize in-task reasoning.

Citation

Luo, Y. T., Liu, T., Pang, P., McKay, D., Chang, S., & Buchanan, G. (2026). Inquiry-based learning patterns in large language model-driven learning environments: an exploratory study from Bloom's perspective. Australasian Journal of Educational Technology, 42(2), 38–57.