A language model writes a product description in a few seconds, but the texts come out similar to each other and unlike the rest of the store's descriptions.
We learn from what you already published
Before the first generation, AutoList reads the descriptions already published in the store and uses them as examples. Along with the request itself, such as "write a description for a fishing rod", the model learns how descriptions in that store are written: how long they are, whether they open with a benefit or a specification, and how technical the tone is.
We fence in what it can invent
A model left free invents categories that do not exist, brands spelled differently and new attributes for every product. AutoList works with closed lists synced from the store: allowed categories, allowed brands, allowed attributes. What is not on the list cannot be chosen.
Keyphrases that do not repeat
If every product gets the same SEO keyphrase, your products compete against each other in Google. We keep track of the phrases already used and refuse a repeat, asking for another.
You approve every product
Nothing publishes itself. Every product lands in an approval screen, with the photos found at the supplier and the proposed copy, and you correct or reject it. It takes longer than automatic publishing, but one wrong sentence published across a thousand products costs more than those few minutes of checking.