How do we ensure that AI-assisted learning leads to conceptual mastery rather than passive reliance? This poster presents a high-impact teaching model from a Physical Chemistry curriculum that integrates Generative AI with real-time classroom polling (Peer Instruction). In this model, polling serves as the "truth mechanism," requiring students to pivot from AI-guided exploration to individual and group accountability.We detail a 6-step instructional workflow where students interact with AI as a Socratic tutor, followed by a rigorous "Human-Only" polling phase to diagnose misconceptions. By leveraging polling data, instructors can immediately identify where AI-driven "guided learning" succeeded or where it led to conceptual errors. The poster showcases visual data on student performance, the "Reveal" strategy for auditing AI hallucinations, and evidence of how polling-driven feedback loops transform the classroom into a laboratory for critical AI evaluation.
