Responsive teaching is an effective teaching approach in which teachers engage and respond to students' mathematical ideas to support their mathematics learning. In this study, the relationship between preservice teachers' (PSTs) noticing expertise and their teaching moves was investigated in a simulated AI chatbot environment. The AI chatbot included a virtual student with misconceptions about fraction operations and was used as an instructional tool to provide opportunities for PSTs to practice responsive teaching skills. Using latent class analysis, PSTs' noticing components (attending, interpreting, and responding) and teaching move patterns were investigated in a sample of 138 PSTs from two universities in Korea. The result indicated that significant correlations between PSTs' noticing expertise and the quality of their teaching moves. In addition, three distinct classes were identified based on PSTs' noticing skills, and differential use of teaching moves was found across these classes. The results underscore the differences in PSTs' noticing expertise in the AI chatbot environment and a typical classroom setting and suggest implications for leveraging AI chatbot simulations in teacher education.