Artificial intelligence (AI) could help predict fashion trends using structured prompts rather than intuition, a new study has concluded. Researchers introduced a Top-Down Prompting technique rooted in the Lotus Blossom method to guide ChatGPT towards specific, consistent forecasts. While outputs often reflect established ideas, the approach also surfaces emerging signals, suggesting a potential complementary tool for expert analysis and wider access for students and small brands.
- The study, conducted by researchers at Pusan National University, compared ChatGPT outputs with professional trend forecasts to assess their accuracy and creativity in predicting men’s fashion for 2024.
- Using Top-Down Prompting, researchers expanded general queries into specific sub-themes like materials, colours, silhouettes, and moods for systematic data generation and analysis.
- ChatGPT mirrored mainstream ideas yet also surfaced new directions, indicating an emerging capacity to detect subtle creative shifts within seasonal menswear forecasting.
- The approach demonstrates how AI can strengthen fashion forecasting by combining computational reach with expert interpretation.
- Authored by Chaehi Ryu and Yoon Kyung Lee from the Department of Clothing and Textiles, the peer-reviewed study in the Clothing and Textiles Research Journal highlights how generative AI is reshaping both fashion forecasting and design education.
THE STUDY: The researchers set out to test whether large language models could forecast fashion trends with measurable accuracy. Their initial trial analysed ChatGPT’s responses to 2023 fall and winter menswear prompts before developing a more structured approach. Their work examined ChatGPT’s ability to interpret cultural and design cues.
- ChatGPT’s predictions were benchmarked against the Official Fashion Trend Information Company fall and winter 2024 menswear forecast to evaluate overlap, specificity, and directional alignment with professional outputs.
- The researchers compared outputs from ChatGPT-3.5 and ChatGPT-4 Classic, assessing whether prompt structure improved consistency, concreteness, and category coverage across silhouettes, materials, key items, moods, and prints.
- Two fashion experts reviewed the analysis, providing an external check on relevance and quality, and helping judge whether the model’s themes matched accepted forecasting standards and practice.
- The project’s stated aim was to complement expert-led analysis rather than replace it, testing whether conversational AI could add structured, accessible support for students and small brands.
THE METHOD: The researchers developed a Top-Down Prompting (TDP) technique, inspired by the Lotus Blossom brainstorming model, to make ChatGPT’s trend forecasts more specific and consistent. This structured process begins with a central query about “Fashion Trends” and expands into sub-prompts covering silhouettes, materials, key items, garment details, decorative elements, colours, moods, and prints, ensuring detailed and repeatable predictions without relying on generic responses or random creativity.
- The Top-Down Prompting framework converts a broad trend query into a mapped hierarchy of focused sub-questions that mirror professional forecasting categories.
- Each layer of the prompt network compels ChatGPT to respond within defined fashion attributes, producing granular and thematically coherent descriptions aligned with industry terminology.
- The Lotus Blossom structure prevents redundancy and improves recall, allowing the model to generate data that can be evaluated quantitatively rather than anecdotally.
- By combining TDP with expert validation, the researchers created a hybrid system that preserves AI’s speed and scope while anchoring its insights in domain-specific structure.
WHAT THE DATA SHOWS: The comparison between ChatGPT’s predictions and the Official Fashion Trend Information Company’s report revealed a mixed picture. The model correctly identified nine of thirty-nine trends, mostly reflecting established styles rather than emerging innovations. Yet it also detected new cultural themes—especially gender fluidity and statement coats—indicating that AI can register subtle creative signals even when quantitative accuracy remains low or inconsistent. However, the authors cautioned that ChatGPT should not yet be viewed as a definitive forecasting tool.
- ChatGPT’s outputs tended to generalise well-known fashion directions, such as oversizing and neutral palettes, instead of producing forward-looking or niche predictions.
- Both ChatGPT-3.5 and ChatGPT-4 Classic repeated several mainstream concepts, confirming that model updates alone do not guarantee more innovative or accurate forecasting.
- The study found that generative AI captured early indicators of social change, suggesting value in exploring its use for detecting evolving aesthetics beyond data repetition.
