Preparing preservice teachers (PSTs) to be able to notice, interpret, respond to and orchestrate student ideas--the core practices of responsive teaching--is a key goal for contemporary science and mathematics teacher education. This mixed-methods study, employing a virtual reality (VR)-supported simulation integrated with artificial intelligence (AI)-powered virtual students, explored the frequent patterns of PSTs' talk moves as they attempted to orchestrate a responsive discussion, as well as the affordances and challenges of leveraging AI-supported virtual simulation to enhance PSTs' responsive teaching skills. Sequential analysis of the talk moves of both PSTs (n = 24) and virtual students indicated that although PSTs did employ responsive talk moves, they encountered difficulties in transitioning from the authoritative, teacher-centred teaching approach to a responsive way of teaching. The qualitative analysis with triangulated dialogue transcripts, observational field notes and semi-structured interviews revealed participants' engagement in (1) orchestrating discussion by leveraging the design features of AI-supported simulation, (2) iterative rehearsals through naturalistic and contextualized interactions and (3) exploring realism and boundaries in AI-powered virtual students. The study findings provide insights into the potential of leveraging AI-supported virtual simulation to improve PSTs' responsive teaching skills. The study also underscores the need for PSTs to engage in well-designed pedagogical practices with adaptive and in situ support.