Generative AI feedback and creative revision in seventh-grade mathematical problem posing

Generative artificial intelligence (GenAI) is entering mathematics classrooms, yet little is known about how middle school students take up AI feedback while posing and revising problems. This classroom-based exploratory study examined feedback uptake, revision depth, and temporal interaction patterns among 40 seventh-grade students from two intact classes. Across two 40–45-minute sessions on fractions and ratio and proportion, students experienced Hint Mode, which provided Socratic questions, and Critique Mode, which provided evaluative comments. Screen-recorded student–AI dialogues, initial and final drafts, and observation notes were analyzed using turn-level qualitative coding, draft comparison, and descriptive pause categories. Across 244 student response turns, acceptance/minimal change was most frequent (51.6%), followed by modification/refinement (31.1%) and conceptual reformulation (17.2%). Reformulation constituted 50.0% of long-pause turns. The findings identify frequent uptake, localized refinement, and occasional conceptual restructuring but do not support comparative effectiveness claims because feedback stance, topic, class, and session order were not separable.