How it works
Content is imported into the base: a sync with the site, files, web pages. It's split into passages and indexed.
When a question comes in, the system finds the relevant passages and passes them to the LLM, which writes the answer from them. That's how RAG (retrieval-augmented generation) works.
The base has to keep up with the brand. If the return policy changes, the base has to change the same day, or the assistant keeps giving the old one. And when the information isn't there, a well-configured assistant says so or hands over, instead of making it up.
Why it matters in e-commerce
Answer quality depends on the base first. A shopper who doesn't get an answer hesitates, and often leaves without buying.
An up-to-date base also takes load off customer service. Questions about shipping, returns or product care get answered without waiting for an advisor.
Further reading: The 6-step checklist to launch your AI shopping assistant
Real-life example
A skincare brand imports its product pages, FAQ and routine guides into the base. A shopper asks whether a serum can be used in the morning with sunscreen. The routine guide covers it, so the assistant gives the brand's recommendation. If no content mentions it, the assistant says so and offers to pass the question to an advisor.
Common confusion
Adding a document to the knowledge base doesn't retrain the model. In most AI assistants, the document is looked up at question time. Changes take effect quickly, and they're just as easy to undo.