Learning to Prompt: Improving Student Engagement with Adaptive LLM-based High-School Tutoring

2026-06-18

AI Summary

This study addresses the limitations of existing static-prompt-based LLM tutoring systems, which struggle to accommodate disciplinary diversity and individual learner differences due to insufficient personalization and dynamic adaptation. To overcome this, the authors propose an adaptive prompt routing mechanism that integrates 14 pedagogical theory-derived features to dynamically select optimal prompting strategies based on subject matter and student learning states. The approach is pretrained in a simulated environment and subsequently deployed in real high school classrooms. Notably, it embeds educational theory features into LLM prompt scheduling for the first time and reveals that stochastic policy sampling significantly outperforms greedy selection. Experimental results show a routing accuracy of 0.694 in simulation (versus baselines of 0.647 and 0.640) and, in real-world settings, an average reduction of three interaction rounds (p = 0.007) along with a 28.1% improvement in exercise completion rates using the stochastic router.

Abstract

LLMs can personalize education, although current static-prompt tutoring systems struggle to adapt to diverse academic disciplines. We develop and test a system with subject-aware prompting, based on 14 pedagogical features (e.g., tutor scaffolding, student understanding) extracted from raw transcripts. We first train a prompt routing model in a simulation environment, and then deploy it for online adaptation with actual high-school students. The simulation benchmark shows the router outperforming two static baselines ($0.694$ vs. $0.647$ and $0.64$, $p<0.001$). A/B testing ($N=656$ conversations from 359 students) shows sim-to-real transfer where the model switches from analytical to scaffolding learning strategies. Our adaptive prompt selection mechanism improves instructional efficiency, maintains pedagogical quality and reduces interactions by around 3 turns ($p=0.007$). While a greedy router achieves a comparable exercise conversion rate with the baseline ($19.1\%$ vs. $19.6\%$), a stochastic router that samples strategies leads to a higher conversion rate ($28.1\%$).

Problem

• adaptive tutoring
• student engagement
• LLM-based education
• prompting
• high-school learning

Innovation

• adaptive prompting
• subject-aware tutoring
• prompt routing
• sim-to-real transfer
• pedagogical features