Answering Student Questions All at Once: Real-Time AI Summaries in Lecture

In large lectures, only a few students typically voice questions, leaving instructors with an incomplete view of what most students are thinking. We evaluated a real-time classroom approach that collects all student questions during class and uses a large language model to group them into 4–7 themes with short, representative questions for the instructor to address. The study spans multiple, large-enrollment undergraduate physics courses and examines three outcomes: (1) how accurately the AI captures the same themes that human experts identify, (2) how students perceive the usefulness of the summaries, and (3) whether seeing the themes helps students ask more applied and integrative questions. Across lectures, the AI summaries closely mirrored the themes generated by experts, and students reported clear benefits: 74% agreed that "seeing what everyone else is thinking" improved their confidence, and 76% enjoyed the AI-generated questions. In a follow-up activity, students' own questions became more focused on application and connections across topics rather than simple repetition. Real-time summarization provides a low-overhead way to make student thinking visible at lecture scale and could help instructors in large courses respond to what the majority of students actually want to know.