RAG (Retrieval-Augmented Generation) first retrieves from a set of documents and then passes the most relevant passages to the model to answer, and Gemini Notebook (formerly Notebook LM) uses this approach to provide answers with citations.