Treffer: Prompting for Engagement: Using the ICAP Framework to Guide Prompt Design in LLM-Powered Dialogue-Based Tutoring System for Novice Programmers.

This paper presents the application of the Interactive–Constructive–Active–Passive (ICAP) framework to analyze novice programmers' engagement in a prototype dialogue-based tutoring system powered by a large language model (LLM). Drawing from annotated dialogue logs, we categorized student utterances as Interactive, Constructive, Active, or Passive. By aligning student responses with tutor scaffolding strategies, we found that Explaining prompts were most effective in eliciting Constructive and Interactive engagement, while Feedback and Instructing often resulted in lower-level responses. We present concrete prompt redesign examples informed by ICAP, illustrating how meta-prompting can scaffold deeper reasoning and co-construction. Although this study does not directly measure learning outcomes, it highlights the value of ICAP as a lens for evaluating engagement in LLM tutors and sets the stage for future work linking engagement to conceptual gains and transfer learning. Our findings offer both theoretical and practical contributions to the design of pedagogically sound LLM-powered tutoring systems. [ABSTRACT FROM AUTHOR]