We developed the AI Math Personalization Tool (AMPT) to explore how personalization and agency over the contexts in math story problems can affect student learning and motivation for mathematics. AMPT uses a conversational interface to learn about students’ interests before creating a math story problem targeting those interests. Feedback gathered through co-design sessions with AMPT provided insight into areas where agency over the story of the problem was particularly important to students (e.g., naming characters) as well as the contextual topics students find interesting. Further, following AMPT sessions, students’ sense of belonging in mathematics increased. By leveraging generative AI to support personalization at scale, AMPT provides a framework for investigating how agency and representation relate to student engagement in mathematics. These insights have broader implications for the design of AI-driven learning tools and the future of personalized education.