Retrieval-augmented generation (RAG) is how an AI answers a question using documents it was never trained on. Instead of relying on what the model memorised, the system first retrieves the most relevant passages from a source you supply — your case files, a research library, a knowledge base — and hands them to the model alongside your question. The model then answers from that text.
RAG is the difference between an AI that confidently guesses and one that can show its work. It is why a tool can cite a paragraph from your contract rather than a generic web summary, and it is the main practical antidote to hallucination.
Why it matters at your desk. For a lawyer, RAG is what lets Harvey answer a question about a specific matter rather than the law in the abstract. For a researcher, it is why Consensus can point you to the actual papers behind a claim, and why Perplexity shows links instead of just prose.
Use RAG when the answer must come from a known, trusted source. It is not magic: if the retrieval step pulls the wrong passage, the answer will be wrong but well-dressed — so the quality of what you feed it still decides the quality of what you get back.