<![CDATA[Generative artificial intelligence (AI) can produce fluent mathematics teaching materials; however, these outputs may contain mathematical errors and misrepresentations of local cultures. This study examined how preservice mathematics teachers engage in truth-seeking when using ChatGPT to design teaching modules that incorporate the local wisdom of Maluku, Indonesia, and investigated the role of a reflective worksheet in supporting this process. Employing the educational design research approach, a worksheet was developed, refined through expert review, and implemented with eight preservice teachers over six sessions. Analysis of chat logs, completed worksheets, modules, and field notes yielded 240 coded pieces of evidence. Only 9% of these fully satisfied the truth-seeking criteria, with an additional 16% reflecting vigilance initiated by the AI tool. Participants identified cultural misrepresentations when personally familiar with the practice but infrequently detected mathematical errors, which were typically flagged by the AI tool. The worksheet facilitated doubt and reconsideration more effectively than it did verification and synthesis, as the proportion of disconfirming evidence increased from 30% to 55%, and all submitted modules retained at least one mathematical error. These findings informed the development of a revised worksheet and seven design principles, summarized as “human first, AI as challenger, and evidence as arbiter” to promote critical and culturally responsive AI use in teacher education.]]>
