RAG (retrieval-augmented generation)

The standard recipe for answers that come from your own documents. A question arrives; the system first searches your files for the passages that cover it, then hands those passages to the AI and says "answer using only this."

An open-book exam instead of a memory test. The clever part isn't the AI — it's that something went and found the right page first.

Three words cover one shelf of ideas, so: RAG is the whole recipe, retrieval is the searching step inside it, and semantic search is the trick that makes the searching work when the wording doesn't match.

Why it matters: RAG is how AI answers from your policies, contracts and case notes — and it does it without anyone rebuilding the AI itself. That distinction is the whole selling point, so it's worth being blunt about the alternative: changing what a model knows from the inside means retraining it, which takes months and serious money and is done by the AI companies, not by you. RAG gets you the same outcome by Friday, by handing it the right pages at the moment of asking. Most "chat with your documents" products are RAG underneath.

Related: Retrieval · Grounding · Semantic search