<p>Generative artificial intelligence tutors have been shown to be highly effective and hold potential to transform education by providing scalable, personalized support. However, little is known about how the system-level prompts that govern AI tutor behavior impact student learning and user experience. In this study, we conducted a randomized controlled experiment using an AI tutoring framework to examine the effects of prompt engineering on students’ perceptions, behavior, and outcomes. Participants (N = 125) were assigned to interact with an AI tutor governed by either a basic prompt—containing minimal instructional guidance—or an enhanced prompt that incorporated established pedagogical best practices, such as active learning, cognitive load management, and growth mindset support. Notably, participants who used the enhanced prompt version engaged in significantly more student-AI interactions, reported higher levels of engagement and motivation, and experienced less cognitive overload. Both groups demonstrated significant learning gains from pre- to post-test. These findings suggest that prompt engineering is a powerful tool for shaping the effectiveness and quality of AI tutoring. Careful design of AI tutor prompts can enhance student engagement and persistence, pointing the way toward more pedagogically responsive and impactful AI-enabled education.</p>