A Study on the Process of Constructing Generative AI Image Prompts Using Hairstyle Search Interest: Focusing on Google Trends Analysis and the Procedural Application of the PDCA Framework

This study aimed to analyze online search interest in hairstyle-related terms and to explore a procedure for converting the results into a data-based structured prompt for generative AI hairstyle image creation. The PDCA (Plan-Do-Check-Act) framework was used as a procedural framework. Google Trends data for seven hairstyle-related terms were analyzed from April 2021 to March 2026, and Sleek Cut, which showed the highest Recent Interest Ratio (2.03), was selected as the case hairstyle. Based on hair design elements and prompt engineering studies, the prompt was structured into Basic, Detailed, and Control elements using R and applied to Google Gemini 20 times to generate 20 images. Three hair and beauty experts independently evaluated the images using five criteria. The overall mean was 4.50, with form showing the highest score (M=4.63) and color the lowest (M=4.15). Average-measure ICCs ranged from .887 to .968, with an overall ICC of .948, indicating high inter-rater agreement. Based on these results, color-related prompt elements were identified as requiring further refinement. The Act stage was limited to deriving improvement directions and did not include regeneration or re-evaluation. These findings suggest that online search-interest data can be used as an exploratory basis for selecting hairstyle visualization targets and systematically structuring prompt elements for generative AI.