Generative artificial intelligence (GenAI) is reshaping how students access and represent disciplinary knowledge, raising ethical and pedagogical questions about cognitive dependence, accuracy, and critical judgement [1]-[3]. This study introduces and examines a prompt-to-image instructional sequence designed to foreground disciplinary reasoning and critical engagement with AI generated visual outputs in undergraduate architectural education. Implemented across six groups, the intervention combined analog reasoning ("brain work") with guided comparison of images generated by multiple AI systems. A total of N = 148 students participated; 125 completed all phases of the activity, and 124 provided valid survey responses.Students first produced hand-drawn visual vocabularies focused on classical architectural orders, typologies, and their defining theoretical features, and later engaged in prompt iteration, cross-AI comparison, and reflective analysis of AI-generated images of the same architectural references. Mixed-methods analysis combining instructor-coded evaluation of image accuracy, qualitative analysis of student reflections, and descriptive survey data suggested recurring conceptual and stylistic inaccuracies in AI-generated images, alongside frequent student recognition of these limitations. The results also highlighted a persistent hyperrealism bias, where visually convincing outputs may blur disciplinary errors.Rather than positioning AI errors as failures, the study suggests how guided visual analysis can frame AI-generated inconsistencies as pedagogically productive moments for reflection and ethical discussion. The proposed model offers a transferable, ethically grounded framework for integrating generative AI into architecture and related STEAM disciplines in a way that foregrounds disciplinary reasoning and ethical reflection.
