Glossary AI technique
RAG (Retrieval-Augmented Generation)
What is RAG (Retrieval-Augmented Generation)?
RAG lets a language model first retrieve the right documents from your own sources and base its answer on those, so it draws on your knowledge instead of only its training.
A language model knows a lot, but nothing specific about your business: not your policy terms, not your case files, not your internal manuals. RAG solves that by first looking up the relevant passages from your sources for every question and passing them to the model. The answer then comes from your knowledge, not from a general memory.
That does two things at once. It makes answers current and company-specific, and it reduces hallucinations, because the model can base itself on real text instead of guessing. You can also trace the answer back to the source, which makes it verifiable.
RAG is often the difference between AI that sounds nice and AI that is right for your situation. For most SMB use cases it is the first building block we reach for, even before we look at fine-tuning.
Last updated: 18 June 2026