Supporting sellers with enhanced product listings

Mercari, the used-goods marketplace that built its business on peer-to-peer selling, has quietly turned into one of the more practical case studies for applying large language models to an unglamorous but very real problem: people are lazy about writing product descriptions. OpenAI's latest customer post describes how Mercari has folded GPT-4o mini and GPT-4 into its listing flow through two features, AI Listing Support and the Mercari AI Assistant, both aimed at getting sellers from a photo and a vague idea to a finished, sellable listing faster.

The mechanics are straightforward. A seller snaps a photo of, say, a secondhand jacket, and the models help generate titles, categories, and descriptions instead of forcing the seller to type all of it from scratch. GPT-4o mini, OpenAI's smaller and cheaper model, handles the bulk of this high-volume, lower-complexity work, while GPT-4 gets called in for tasks that need more nuance. That kind of tiered model deployment is becoming the standard playbook for companies bolting AI onto existing products: use the expensive model sparingly, lean on the cheap one for everything else.

What makes this notable isn't the novelty of AI-written product copy, plenty of e-commerce tools already do that, but the scale and the specific friction point Mercari is targeting. Marketplaces live and die on how quickly a seller can list an item and how appealing that listing looks to a buyer scrolling through hundreds of similar posts. Anything that shortens the gap between