I think the example you give is a little backwards — a RAG system searches for relevant content before sending anything to the LLM, and includes any content retrieved this way in the generative prompt. User query -> search -> results -> user query + search results passed in same context to LLM.