How Etsy Uses Gemini and Vertex AI to Personalize 90 Million Shopping Experiences

Etsy exists to make human-to-human commerce work at scale. Every item on the platform is made, handpicked, or designed by a real person, and the marketplace’s core promise is that buyers can find something genuinely special rather than mass-produced. With more than 130 million listings from 5 million sellers and a buyer base approaching 90 million people, that promise becomes an engineering challenge: how do you surface the right item for each specific buyer out of a constantly changing, enormously varied inventory?

The challenge was compounded by the nature of the inventory itself. Unlike commodity e-commerce, Etsy’s listings are irregular in taxonomy, metadata quality, and categorization. A shopper searching for a “vintage hat” might mean any number of things, and the seller who listed it may not have described it in terms that match the buyer’s intent. Traditional keyword matching left too much discovery potential on the table. Etsy needed a way to enrich its understanding of inventory at the scale of 130 million items — and to do it continuously as new listings arrived daily.

Etsy built a data and AI platform centered on BigQuery as the warehouse, Dataflow for ingestion pipelines, and Vertex AI as the unified surface for machine learning and generative AI. Gemini models form the core of the enrichment and personalization layer. The team starts by building a foundational metadata dataset using Gemini, connecting, verifying, and validating listings with attributes that go beyond what sellers provide. Gemini also helps Etsy’s merchandising team identify emerging cultural trends from social media and rapidly classify related new listings before the trend has fully formed.

For buyer discovery, Etsy uses what it calls “algotorial curation” — a combination of human editorial curation and AI amplification. Merchandising experts create seed collections that exemplify a trend or aesthetic. Vertex AI then finds semantically similar listings and scales each collection by 80x, so what a team of experts can manually curate becomes a personalized discovery surface for every individual buyer. The Home feed shows trend collections personalized to each shopper’s taste and price point. On the SEO side, Gemini-powered alt text generation for listings increased search engine-driven visits by 5% and boosted seller conversions by 3%.

Etsy describes this shift as the foundation for a new way to “discover special.” The platform is moving into multimodal AI, using images, videos, and text together to understand both inventory and buyer intent at a richer level. For sellers, the benefits compound as well: better metadata and richer discovery surfaces mean their items reach buyers more efficiently, without requiring sellers themselves to become SEO experts.