Artificial Intelligence (AI) technology has been widely used in architectural design. However, many architects remain skeptical about AI-generated design schemes. This study aims to explore the reasons behind architects’ dissatisfaction with the AI-generated design. We invited professional and non-professional participants to evaluate three hospital design schemes: two generated by AI application EvoMass and one by experienced architects. Visual behavior data and oral description of participants during the evaluation were collected and analyzed. The results show significant differences in focal points and observation patterns. Professional participants paid more attention to the building footprint and the junction of buildings and city roads. These findings indicate that understanding of architectural design principles is crucial to enhancing Generative AI’s capabilities.