Supplementary Materials for Anand, S et al., 2025.

This study evaluates the effectiveness of large language models (LLMs), specifically Claude Sonnet 4.0 and ChatGPT 4.1, for analyzing formative feedback to support student-centered learning (SCL). In large courses, instructors often struggle to promptly review and synthesize student input. We used a low-stakes task: analyzing 63 anonymous student responses to a Muddiest Point prompt in a human physiology class after a lecture on respiratory physiology. Across 20 runs, both LLMs consistently identified “Ventilation and Lung Mechanics” as the most frequent area of confusion, aligning with human analysis. LLMs completed the task significantly faster than a human reviewer (average 19.6 s/31.0 s vs. 32 min), with thematic reliability. This suggests LLMs can efficiently generate information from student input, enabling instructors to adapt their instruction in real-time. The approach supports SCL and educational equity through inclusion of all student voices. While promising for formative feedback, observed variability indicates that further refinement is needed before LLMs are utilized for high-stakes summative assessment.