Picture a flea market the size of a small country. Every stall is run by a different person. Every item is one of a kind. Now imagine walking in and knowing, within seconds, exactly where to go. That is what Etsy is building. And the business case for getting it right is enormous.
Etsy’s buyers now number roughly 90 million. Most of what they are looking for is handmade, vintage or customizable and therefore unlike anything a standard search engine was built to retrieve. Etsy’s engineering team said the old discovery infrastructure was not breaking down slowly. It had already broken.
Using Google Cloud’s Gemini models, Vertex AI and BigQuery, Etsy rebuilt its search and recommendation system to read every listing’s images, text and contextual signals simultaneously. The platform now categorizes more than 130 million items from over 5 million sellers dynamically, personalizes each buyer’s home feed in real time and tracks cultural trends as they emerge on social media before they peak.
The early results are measurable: Etsy recorded its first sequential gain in active buyers in eight quarters, PYMNTS reported.
When the Catalog Became the Problem
Search and recommendation systems built for standardized retail shelves struggle when every item is one of a kind. A shopper looking for a “rustic anniversary gift with a botanical feel” cannot generate a clean keyword. A static merchandising team cannot manually categorize 130 million listings at the rate sellers add them.
Etsy told Google Cloud that the company needed generative artificial intelligence (AI) to do what was not economically or logistically possible before large language models existed.
Advertisement: Scroll to Continue
The Algotorial Curation Model
Etsy calls its hybrid approach algotorial curation. Human merchandising experts create seed collections built around themes and styles. Generative AI then scales those seeds across the full catalog, finding semantically similar listings the team never would have reached manually. Listings per theme increased roughly 80 times, Google Cloud reported. Improved alt text generation using Gemini lifted SEO-driven visits 5% and boosted seller conversions 3%.
The system also reorders what each buyer sees. Which trends appear on a shopper’s home feed, and in what sequence, is specific to that individual’s profile. Past behavior, favorited items and browsing patterns feed a model that continuously adjusts the experience in real time.
Buyer Behavior Is Starting to Move
The business signal came in Q1 2026. Active buyers reached 86.6 million, down year over year but up sequentially for the first time in eight quarters, PYMNTS reported. Mobile app sales, where the new personalization model runs most thoroughly, grew 11.2% year over year, up from 6.6% the prior quarter.
Etsy is also expanding into conversational search. An Etsy app in ChatGPT is now live in beta, letting users type prompts and receive relevant listings directly inside the conversation, PYMNTS reported. The company described early agentic traffic as high-intent. It remains a fraction of total visits for now. Etsy’s bet is that it will not stay that way.
