Individual differences in students’ learning preferences highlight the need to move beyond uniform instructional models toward personalized and adaptive education. This need is particularly important in architectural education because of its visual, process-oriented, and project-based nature, which integrates analysis, design, critique, and practical experience. This study aimed to design and validate a framework for adaptive architectural education based on VARK learning preferences and supported by Large Language Models (LLMs). The study was applied in purpose and descriptive–analytical and design-oriented in approach, combining survey research, conceptual analysis, framework design, and validation. The study population comprised undergraduate architecture students at the Faculty of Art and Architecture, University of Guilan, with 102 students assessed using the VARK questionnaire. Results showed that the visual preference had the highest frequency (43.1%), followed by the multimodal preference (23.5%). The temporal reliability of the VARK questionnaire, measured using Cohen’s kappa, was 0.71, while the internal reliability of the educational content evaluation instrument, measured using Cronbach’s alpha, was 0.91. Based on the learning profiles, educational requirements and adaptive prompt rules were developed, and prototype content on “spatial hierarchy in architectural design” was generated using LLMs. Evaluation by three experts yielded a mean score of 4.43 of 5, while evaluation by 40 students yielded 4.28 of 5. Among students, the highest mean score concerned the relevance of content to architectural education (4.40), whereas the lowest concerned interactivity (4.12). The findings indicate that VARK profiles can be systematically translated into educational needs and strategies and subsequently into prompt-design rules, enabling differentiated presentation of educational content. Accordingly, the study provides a traceable framework linking learning preferences, instructional decision-making, prompt design, and adaptive content generation in architectural education. However, the findings demonstrate the framework’s initial suitability and feasibility rather than its effectiveness in improving actual learning outcomes.