Integrating ChatGPT in Mathematics Teacher Education: AI-Based Simulation Role-Playing to Support Practice-based Teaching

Synthesis: This study designed Student GPT, a custom ChatGPT-based chatbot that role-plays a middle school student holding common misconceptions in ratio reasoning, giving preservice secondary mathematics teachers low-risk, personalized practice-based teaching experience in mathematics education. Analyzing teacher–bot chat histories through an inductive qualitative coding framework (the Affective, Communicative, Technical "ACT" framework), the authors found the simulated student performed well on clarity, relevance, error types, knowledge, and consistency, but struggled with authentic student tone and role confusion. The work demonstrates how Generative AI can power Simulation role-playing to support Teacher Education practice-based teaching.

Key Findings

• Researchers built a customized AI-based role-play simulation using the OpenAI custom-GPT feature, prompting it to emulate a middle schooler struggling with ratio concepts; a refined prompt (Prompt B) that specified three literature-grounded misconceptions — confusing additive and multiplicative reasoning, lacking covariational thinking, and over-relying on a single strategy — far more reliably elicited conceptual errors than a broad algebra prompt (misconception presence 0.98 vs. 0.40; χ²(1, N=200)=78.64, p<.001).
• Five preservice secondary mathematics teachers interacted individually with Student GPT on the Orange Juice Problem in a methods course, diagnosing the simulated student's misconceptions and guiding it toward correct ratio solutions.
• Using an inductive qualitative content analysis of the chat histories, the authors developed and applied an original Affective, Communicative, Technical (ACT) coding framework for systematically assessing the chatbot's role-play simulation behaviors, strengths, and weaknesses.
• In the affective domain, Student GPT reliably used positive, accepting, and encouraging language and stayed on task, but also voiced negative attitudes toward mathematics ("Math can be tricky") that preservice teachers sometimes echoed, raising concerns about impacts on their mathematical identity.
• In the communicative domain, responses were clear and topically relevant, but the simulated student's language was often more teacher-like than student-like — formal, compound-complex, redundant, and prone to restating — reducing the authenticity of the simulated conversational exchanges.
• In the technical domain, the bot correctly embodied the three target misconceptions and age-appropriate knowledge, but it learned new strategies unrealistically fast (grasping approaches in a single turn) while inconsistently struggling with a common-denominator strategy, and it periodically switched roles between struggling student, competent student, and teacher (role confusion).
• The authors conclude that despite these authenticity limitations, Student GPT is an affordable, easily customizable, content-specific complement to costly platforms (e.g., TeachLivE™) for building preservice teachers' pedagogical content knowledge about student misconceptions, aligning with the decomposition and "approximations of practice" from practice-based theory of teacher learning.

Connected Concepts

• Teacher Education
• Math Education
• Simulation
• Generative AI
• Teacher Role
• Pedagogical LLM Training
• Professional Training

Connected Articles

• LLM Student Simulation Teacher Insights
• Simulating Students LLM Review 2026
• Valid Student Simulation LLM 2026
• AI TPACK Preservice Math Teachers
• Mathematics Teachers Chatbot Motivation 2026

Citation

Zhuang, Y., & Zhang, S. (2025). Integrating ChatGPT in Mathematics Teacher Education: AI-Based Simulation Role-Playing to Support Practice-based Teaching. International Journal of Artificial Intelligence in Education, 35, 3873–3895.