<p>This study compares the visual representations in Vietnamese and Taiwanese science textbooks and evaluates the feasibility of using generative artificial intelligence (ChatGPT) as a coding assistant for image analysis. Grounded in Grammar of Visual Design (Kress &amp; van Leeuwen, 2020), visual conventions, and human–artificial intelligence (AI) collaborative research methods, the study adopted a dual approach: manual content analysis of biological images focusing on classification structures and AI-assisted coding in which ChatGPT was guided by the same coding scheme and decision rules. Images were sampled from Grade 6 (Vietnam) and Grade 7 (Taiwan) chapters on biological classification. Manual analysis revealed that textbooks in both countries' predominantly used covert visual designs which mainly composed of photographic contents to convey classification concepts. Vietnamese textbooks exhibited a more balanced distribution between organism identification and classification, whereas Taiwanese textbooks placed greater emphasis on the implicit presentation of classification. ChatGPT’s codes showed considerable agreement with the human analysis, particularly in recognizing basic analytical images and taxonomic structures. However, the model was less reliable in decoding nuanced visuals, implicit relations between captions and images, and unconventional layouts. These findings suggest that AI can serve as an assistant to improve the efficiency and consistency of visual data analysis, while human oversight remains essential for context-sensitive interpretation and theoretical alignment. The study contributes to human–AI collaborative research in education and offers practical insights for textbook developers and science educators in cross-cultural contexts.</p>