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

Etsy, the global marketplace for handcrafted and vintage goods, serves nearly 90 million buyers across more than 130 million listings from 5 million sellers. Using Vertex AI, BigQuery, Dataflow, and Gemini, the company built a personalized search and discovery platform it calls “algotorial curation” — increasing listings per theme by 80x, driving a 5% lift in SEO-driven visits, and delivering a 3% conversion improvement for sellers.

Impact

~80x

Listings per theme increase via algotorial curation

5%

SEO-driven visits increase

3%

Seller conversion lift from SEO optimization

Nearly 90 million

Buyers served with personalized experiences

Challenge

With 130 million listings from 5 million sellers, Etsy needed a way to deeply understand its constantly changing inventory, determine individual buyer intent, and deliver personalized discovery experiences at scale — something that traditional keyword-based approaches and manual curation could not accomplish.

Solution

Etsy deployed Gemini, Vertex AI, BigQuery, and Dataflow to enrich listing metadata at scale, amplify human-curated collections by 80x through semantic similarity models, and deliver individualized buyer feeds and search experiences across nearly 90 million shoppers.

Tools & Technologies

What Leaders Say

There’s only one marketplace where you know you’re buying from another real person on the other side, and that’s Etsy. We’re not trying to replace that human connection. We want to amplify and scale that connection with AI.

Mahir Yavuz, Senior Director of Engineering for Search, Recommendations, and Ads, Etsy

Our inventory comprises more than 130 million items from more than 5 million sellers. Google gen AI and LLM technologies help us enrich our data to better understand our inventory’s distinctive characteristics.

Mahir Yavuz, Senior Director of Engineering for Search, Recommendations, and Ads, Etsy
Get the full story.

Sign up to read complete case studies, access detailed metrics, and unlock all use cases.

Full Story

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.

Similar Cases

M
Morrisons
98.96%
data reporting lag reduction

Morrisons, one of the UK’s largest supermarkets serving nine million customers weekly across 500 stores, migrated its on-premise data warehouse to BigQuery and Looker, reducing reporting lag by 98.96% from one day to 15 minutes. Real-time data now powers Vertex AI demand forecasting models and a customer-facing Product Finder app that receives 50,000 hits per day during peak periods.

RetailBBigQueryVAVertex AI
CE
Class Editori
One of Italy’s first AI agents in the media landscape
mfgpt launch milestone

Class Editori, a leading Italian media company specializing in finance, fashion, and lifestyle with 40 years of content archives, partnered with Softlab to build MFGPT on Google Cloud — one of Italy’s first generative AI agents in the media industry. The system unified four decades of journalistic archives and real-time financial data into BigQuery, powered by Gemini and Vertex AI, converting trial users into paid subscribers and securing B2B enterprise agreements with major financial institutions.

Advertising & MediaBBigQueryVAVertex AI
FD
Fifth Dimension
Days or weeks → 30 minutes
investment memo drafting time

Fifth Dimension, a global AI platform for commercial real estate asset managers and owner-operators, built a multi-model workflow on Google Cloud using Gemini for large-scale document ingestion and Claude for high-precision reasoning. The platform compressed investment memo drafting from days or weeks to just 30 minutes and achieved 99.9% reliability for multi-hour workflows, driving deals with top-10 U.S. asset managers.

Real EstateVAVertex AIBBigQuery
AS
AXA Switzerland
Over 95%
query and processing time reduction

AXA Switzerland, the country’s leading insurer covering over 40% of Swiss companies, migrated its entire data infrastructure to Google Cloud and deployed BigQuery, Vertex AI, and Gemini to become a data-driven organization. The transformation reduced complex query times from days to minutes or seconds and generated a high double-digit million Swiss franc profit improvement through Smart Data initiatives.

InsuranceBBigQueryVAVertex AI
GA
Giles AI
95%
medical research data extraction accuracy

Giles AI, a London-based healthcare AI startup, built its medical research assistant on Google Cloud using Vertex AI, Gemini Pro, and Document AI to help researchers extract structured insights from millions of scientific articles. The platform achieved 95% accuracy in data extraction, a 98% agreement rate with human researchers, and helped one clinical customer cut research task time by 85%.

HealthcareVAVertex AIGGemini
FD
Fifth Dimension
50x
document processing capacity increase

Fifth Dimension, a UK-based AI analytics company serving the real estate industry, migrated to Google Cloud to overcome critical infrastructure bottlenecks. By adopting Vertex AI, Cloud Run, and serverless architecture, the company achieved 50x processing scalability, 6x revenue growth, and a 30% reduction in infrastructure costs — all within a rapid growth trajectory from founding in 2023 to global scale by 2025.

TechnologyVAVertex AICRCloud Run
GF
Grupo Falabella
3x
growth in whatsapp conversations

Grupo Falabella, one of Latin America's largest retailers, deployed Salesforce Agentforce on WhatsApp to autonomously handle customer service inquiries across seven countries. The AI agent resolves 60% of requests without human intervention, operates 24/7, and has driven a 3x increase in WhatsApp conversations in just three months — shifting customer support away from costly phone channels.

RetailAAgentforceA3Agentforce 360 Platform
R
Rakuten
~50% reduction
mean time to recovery

Rakuten integrated OpenAI Codex into incident response, CI/CD pipelines, and autonomous development — cutting mean time to recovery by 50% and compressing quarter-long projects into weeks.

RetailOCOpenAI Codex