<p>The complexity of programming instruction, coupled with individual differences in student ability, poses challenges to traditional lecture-based teaching. While the flipped classroom model enhances interaction, limited instructional resources often fail to address diverse, real-time learning needs. To address this gap, this study investigates the effectiveness of integrating “differentiated chatbots” into a flipped classroom and identifies their advantages over conventional QA chatbots. Participants were divided into experimental and control groups; the former utilized a differentiated chatbot framework based on cognitive apprenticeship (comprising Tutor-bot and Tutee-bot), while the latter used a standard QA chatbot. Unlike ChatGPT, the differentiated chatbot actively scaffolds students’ thinking and elicits responses rather than passively providing answers. The results demonstrate that the experimental group significantly outperformed the control group in learning effectiveness, motivation, self-efficacy, and the perceived “usefulness” and “necessity” of programming (with no significant differences in confidence or liking). This study contributes to the field by empirically demonstrating that educational chatbots incorporating differentiated instruction and role-playing mechanisms can effectively overcome the constraints of large-class sizes and instructor availability. Furthermore, it provides a theoretically grounded and viable framework for personalized programming education.</p>