The ability of an artificial intelligence tool to recognise types of triangles in mathematics: an analysis based on the Toulmin model

Introduction This study explores the structural characteristics and reasoning patterns of an artificial intelligence tool (Google Gemini 1.5 Flash) acting as a simulated student in middle-school geometry argumentation. Methods A qualitative single-case study design using directed qualitative content analysis was employed. The dialogue sequences were operationalized and coded using Toulmin's model of argumentation. Results The findings demonstrate that while the AI tool effectively generates rule-based warrants and proof schemes, it exhibits limited spontaneous inclusion of Data, Rebuttal, and Qualifier components, frequently requiring structured teacher interventions to overcome repetitive justification loops. Discussion The study underscores the vital role of intentional teacher scaffolding in AI-supported instructional environments and highlights the utility of AI tools as simulated discussion partners for teacher preparation and task design.