RAG

RAG is a pattern where a language model retrieves relevant documents first and generates an answer grounded in them, rather than answering from its training data alone.

Also known as: retrieval-augmented generation

Retrieval supplies facts the model does not reliably hold: current prices, this site's policies, this customer's order. Generation supplies fluent synthesis. The quality ceiling is set by retrieval — a well-written answer built on the wrong documents is worse than no answer, because it is convincing.

On a website, RAG is the right shape for factual and support questions and the wrong shape for browsing. Nobody wants a paragraph describing six jackets; they want to see the jackets. Routing by intent before choosing between generation and results is what separates useful implementations from demos.