Impact of Artificial Intelligence Generated Feedback on Math Quiz Scores

Abstract This study examined the effect of feedback on students' responses to quiz questions generated by artificial intelligence (AI), specifically using ChatGPT-3.5, on student math quiz performance at a private school in the Philippines. Immediate, personalized feedback is known to support effective learning, yet teachers often struggle to provide it consistently due to large class sizes and limited teaching time. Over 14 weeks, 170 students from Grades 1 to 10 participated in this study, completing 1,348 quiz response sets using the school's existing online learning system, the Moodle Learning Management System (LMS). Students were randomly assigned to two groups: one group received basic correctness-only feedback, while the other received detailed, AI-generated feedback that included explanations, hints, and encouragement tailored to their responses. Students scoring below 80% on initial quizzes received feedback from their group before retaking similar quizzes. Those who received detailed AI-generated feedback improved significantly more (σ = 1.1, Hake's g = 21.6%) than the correctness-only group (σ = 0.7, Hake's g = 13.8%), a statistically significant difference (p < .01, Cohen's d = 0.40). These results indicate that detailed AI-generated feedback effectively supports math learning, providing scalable, personalized instructional assistance within everyday classroom environments across diverse grade levels.