Custom Generative Artificial Intelligence Tutors in Action: An Experimental Evaluation of Prompt Strategies in STEM Education

The integration of generative artificial intelligence, particularly large language models, into education presents opportunities for both personalised learning and pedagogical challenges. This study focuses on electrical engineering laboratory education. We developed a configurable prototype of a generative artificial intelligence powered tutoring tool, implemented it in an undergraduate electrical engineering laboratory course, and analysed 208 student–tutoring tool interactions using a mixed-methods approach that combined research team evaluation with learner feedback. The findings show that student prompts were predominantly procedural or factual, with limited conceptual or metacognitive engagement. Structured prompt styles produced clearer and more coherent responses and were rated the highest by students, while approaches aimed at fostering reasoning and reflection were valued mainly by the research team for their pedagogical depth. This contrast highlights a consistent preference–pedagogy gap, indicating the need to embed stronger instructional guidance into artificial intelligence tutoring. To bridge this gap, a promising direction is the development of pedagogically enriched AI tutors that integrate features such as adaptive prompting, hybrid strategy blending, and retrieval-augmented feedback to balance clarity, engagement, and depth. The results provide practical and conceptual value relevant to educators, developers, and researchers interested in artificial intelligence tutors that are both engaging and pedagogically sound. For educators, the study clarifies how students interact with tutors, helping align artificial intelligence use with instructional goals. For developers, it highlights the importance of designing systems that combine usability with educational value. For researchers, the findings identify directions for further study on how design choices in artificial intelligence tutoring affect learning processes and pedagogical alignment across STEM contexts. On a broader level, the study contributes to a more transparent, equitable, and sustainable integration of generative AI in education.