Abstract This study investigated the instructional effects of four ChatGPT-augmented feedback strategies in a Jupyter-based programming learning environment. Based on whether students received “error-correction hints” and/or full “solution guidance”, we designed four distinct feedback conditions and implemented a quasi-experimental study with undergraduates from a university in Sichuan, China. Learning achievement, engagement, and cognitive load were compared across the four groups. Results indicated that combined hints-plus-solutions feedback best sustained improvements on immediate tests, whereas hints-only feedback yielded superior transfer-test performance, likely because learners engaged more deeply in problem solving when not given full solutions. No significant differences in learning engagement emerged among the four conditions, suggesting that all feedback types similarly motivated students. Importantly, the hints-only group experienced a moderate cognitive load that supported strong transfer outcomes, underscoring the need to balance informational support and mental effort in feedback design. Survey data further revealed generally positive student attitudes toward ChatGPT feedback, tempered by concerns about accuracy and overreliance. These findings suggest that adaptive use of AI-powered hints can foster both skill consolidation and autonomous problem-solving, and they highlight the importance of monitoring cognitive load. Implications for practice include guiding students to critically interpret AI feedback, integrating teacher-led case analyses to deepen conceptual understanding, and calibrating feedback depth to optimize cognitive challenge. This study contributes empirical evidence on ChatGPT’s role in programming education and offers actionable strategies for blending AI hints and solutions in instructional design.
Xiao et al. (Thu,) studied this question.