sweetmelon
← Back to Retail & Consumer
MarketingRetail & ConsumerChatGPTClaudePosted

An AI that localizes product pages from photos and three layers of rules

sweetmelon team · Marketing

Your store sells products from overseas suppliers, and every listing arrives in another language. Hundreds of pages need to read naturally in yours, and machine translation gives you stiff copy that no shopper would type into a search bar. Give an AI each product photo with the original title and description, plus three layers of rules. General rules set how a title is built and how long a description runs, category rules say which details matter (weight, for running shoes), and brand rules set the tone. Then add rules for mistakes that only show up at volume. Words printed on a shirt stay out of the model name. The category gets narrowed to what local shoppers search for, so a sandal with a tall heel becomes a high-heeled sandal. And if a logo shows in the photo but the supplier never mentioned it, the copy should. Pages come out sounding much more local. Before publishing a whole category, read a handful of its pages yourself, because one bad rule repeats in every item it touches.

One vote per person.

Comments

Tried this at work? Add what happened.

Sign in to comment