Glossary/Retrieval-augmented generation
Retrieval-augmented generation
Also known as: RAG
Definition
Retrieval-augmented generation (RAG) is an architecture where a model retrieves relevant documents at query time and uses them to generate a grounded, citable answer.
RAG pairs a language model with a search step. When a question arrives, the system first retrieves documents that relate to it, then feeds those documents to the model as context so its answer reflects the retrieved material rather than memory alone.
It is the engine behind most AI search products. Perplexity, Google AI Overviews, and assistant browsing modes all follow this pattern: search, retrieve, then generate. The quality and selection of retrieved sources shape what the final answer says and cites.
For a firm, RAG is why external visibility matters. If a firm's pages and mentions are among the documents an engine can retrieve for a relevant question, they can end up inside the generated answer.