#489: AI-Scaffolded Summarization in STEM: How do Students Differ in Writing and Revision?

Summarization is a key learning strategy in undergraduate STEM, yet students often struggle without timely, effective feedback. This study investigates how an AI-scaffolded feedback system (SMART) relates to students’ revision behaviors and how these patterns vary across diverse student backgrounds. We analyzed 1,081 summary revisions from 49 undergraduates in an introductory physics course using a suite of mixed-effects models. Results show that adding concepts and words while maintaining semantic coherence predicted stronger concept learning, whereas excessive deletion corresponded to weaker outcomes. Revision patterns differed meaningfully by gender, race/ethnicity, and academic level, suggesting that AI-scaffolded feedback needs to be tailored to support diverse learners toward productive revision strategies.