How adding metacognitive requirements in support of AI feedback in practice exams transforms student learning behaviors

Texto completo

Providing personalized, detailed feedback at scale in large undergraduate STEM courses remains a persistent challenge. We present an empirically evaluated practice exam system that integrates AI generated feedback with targeted textbook references, deployed in a large introductory biology course. Our system specifically aims to encourage metacognitive behavior by asking students to explain their answers and declare their confidence. It uses OpenAI's GPT-4o to generate personalized feedback based on this information, while directing them to relevant textbook sections. Through detailed interaction logs from consenting participants across three midterms (541, 342, and 413 students respectively), totaling 28,313 question-student interactions across 146 learning objectives, along with 279 post-exam surveys and 23 semi-structured interviews, we examined the system's impact on learning outcomes and student engagement. Analysis showed that across all midterms, the different feedback types showed no statistically significant differences in performance, though there were some trends suggesting potential benefits worth further investigation. The system's most substantial impact emerged through its required confidence ratings and explanations, which students reported transferring to their actual exam strategies. Approximately 40% of students engaged with textbook references when prompted by feedback---significantly higher than traditional reading compliance rates. Survey data revealed high student satisfaction (M=4.1/5), with 82.1% reporting increased confidence on midterm topics they had practiced, and 73.4% indicating they could recall and apply specific concepts from practice sessions. Our findings demonstrate how thoughtfully designed AI-enhanced systems can scale formative assessment while promoting sustainable study practices and self-regulated learning behaviors, suggesting that embedding structured reflection requirements may be more impactful than sophisticated feedback mechanisms.