Abstract
High-quality physics instruction depends not only on what teachers know but also on how they can translate disciplinary ideas into intelligible, responsive explanations for learners who struggle. However, student-teachers often have limited opportunities for repeated, low-stakes rehearsal of dialogic explanations that target misconceptions in real time. This study, which was conducted in Kazakhstan, examined whether sustained weekly practice with a generative artificial intelligence (GAI) chatbot built on the Llama 3.1 model, simulated struggling secondary school students, and augmented preservice physics teachers’ professional development across one academic semester. A nonequivalent quasiexperimental pretest–posttest design with historical controls was implemented in an undergraduate physics teacher education program. The participants (n = 86) either received access to a purpose-built Telegram chatbot (n = 41) or completed the same semester without chatbot access (n = 45). Outcomes were assessed via an externally coded observational measure of explaining performance, a standardized test of physics-related pedagogical content knowledge (PCK), and a physics teaching self-efficacy scale. The findings indicated clear advantages for the chatbot condition in explaining performance (d = 0.90), suggesting that repeated, misconception-centered dialog with a simulated student can sharpen enacted explanatory practice. In contrast, differences favoring the chatbot group in terms of PCK (d = 0.35) and physics teaching self-efficacy (d = 0.20) were modest and not conclusive within the semester timeframe. The study contributes early quantitative evidence that scalable AI-driven student-persona simulations can bolster practice-near instructional communication while also delineating boundaries of impact for broader knowledge and belief outcomes. The results support the positioning of chatbot-based rehearsal as a targeted supplement within physics teacher education.