<p>Emotions play a central role in how students engage with mathematical problem-solving, yet teachers often have limited access to students’ moment-to-moment emotional experiences, particularly during independent technology-supported work. This study examines how AI-generated emotional alerts are incorporated into mathematics teachers’ real-time instructional decision-making and how teachers’ responses to these alerts participate in students’ engagement during mathematical problem-solving. Drawing on Goldin’s framework of engagement structures, which considers affect as an integral component of cognition, a qualitative case study was conducted with one teacher and eight high-achieving high school students using Dash4Emotion, an AI-generated emotional alert system. Data sources included video recordings of classroom interactions, logs of the AI-generated emotional system, and stimulus recall interviews. The findings identified five recurring patterns through which the alerts became part of the teacher’s instructional decision-making. These patterns show how the alerts directed the teacher’s attention to particular students or moments and were interpreted alongside students’ observable actions and the mathematical demands of the task. The paper argues that the pedagogical significance of AI-generated emotional alerts lies in how these alerts reshape what teachers observe and how they respond to students during ongoing mathematical problem-solving. This study contributes to research on AI in mathematics education by conceptualizing affective AI systems as mediating resources within teachers’ instructional decision-making rather than as autonomous instructional agents.</p>