AI-driven granular LCA for apparel: Quantifying product-specific GHG emissions for sustainable design and circularity

Product-specific Life Cycle Assessment (LCA) for apparel is hindered by generalized data, failing to quantify unique environmental burdens crucial for sustainable production and consumption. This study addresses this challenge by proposing and validating an automated methodology using a Vision-Language Model (VLM) to extract product-specific details (e.g., fabric composition) from unstructured User-Generated Content (UGC) on consumer-to-consumer (C2C) platforms and subsequently calculate corresponding product-specific greenhouse gas (GHG) emissions. The study seeks to demonstrate the significance of AI-enabled product-specific information as a tool for deriving insights into sustainable production, cleaner manufacturing, and resource efficiency. Applying this framework to 3500 apparel listings from a major C2C marketplace, our VLM demonstrated high accuracy in extracting physical attributes like garment size (92.3%) and fabric composition (81.6%). The subsequent LCA quantified an average emission of 11.9 kg-CO 2 eq/item. More importantly, by analyzing the product-related parameters required for these computations, we attempted to extract and structure data on product attributes and user behavior from unstructured UGC, enabling more effective estimation of product-specific GHG emissions. Although challenges remain in precisely inferring complex user behaviors from unstructured data, the successful granularization of material composition highlights the significant impact of material selection, providing actionable insights for sustainable eco-design and circularity. Ultimately, by significantly enhancing information transparency across the value chain, our work directly aligns with emerging global policies like the EU Digital Product Passport (DPP), fostering data-driven decision-making for cleaner production goals.