Etsy uses large language models to automatically read seller listing titles, descriptions, and photos and pull out structured product details — like size, color, material, and style — that sellers did not explicitly fill in. This expanded complete attribute coverage in key categories from 31% of listings to 91%.

Because Etsy's 100+ million listings are not linked to a standard product catalog or manufacturer database, structured product data is sparse and inconsistent. Etsy uses large language models with context engineering — providing seller-supplied data as structured JSON input — to infer missing attributes in parallel across many listings at once. The extracted attributes are used to power search filters, color swatches on results pages, and recommendation quality. Separately, Etsy uses visual representation learning with deep learning models (EfficientNetB0 and EfficientFormer) trained on product images to improve ad targeting, search, and recommendations, yielding measurable improvements in click-through rates and purchase rates in sponsored search results.