<![CDATA[This study aims to analyze the quality of mathematical proofs constructed by prospective mathematics teachers at UIN Datokarama Palu with ChatGPT's assistance and to describe patterns of ChatGPT use during the proof-construction process. The study employed a qualitative case study design. The participants were 12 students who were given a proof problem and interacted with ChatGPT three times on different days before independently writing their final answers. Data were collected through a proof test, documentation of interactions with ChatGPT, and interviews, and were analyzed thematically using Stylianides’ three criteria for mathematical proof and the seven stages of generative AI interaction proposed by Yoon et al. The findings showed that 11 of the 12 students constructed proof that met Stylianides’ criteria for valid mathematical proof, with the main variation concerning the explicitness of mathematical rules. One student demonstrated an error in the modes of argumentation criterion. Regarding the patterns of ChatGPT usage, all prospective teachers followed only six stages based on the framework proposed by Yoon et al.; however, they consistently failed to perform verification or further exploration when it came to determining how to continue using ChatGPT. These findings suggest that students primarily used ChatGPT as a source of quick answers rather than as a reflective dialogic partner, highlighting the need to strengthen critical verification in AI-assisted mathematical proof learning]]>
