Glossary
RAG (retrieval-augmented generation)
RAG (retrieval-augmented generation) means the AI first looks things up in a trusted source (your documents, a database, the web) and then writes its answer from what it found. It cuts hallucinations and keeps answers current.
An LLM on its own answers from memory: everything it absorbed during training, frozen at some past date. RAG adds an open-book step: before answering, the system retrieves the relevant pages from a chosen source and hands them to the model along with your question.
Picture a pharmacist. Asked about a rare interaction, a good one doesn’t recite from memory. They pull up the official database, read, then answer. Same competence, but grounded in a source that’s current and checkable.
For you this means two things. Answers can cite documents the model never saw in training (your contracts, today’s news, a product manual). And hallucinations drop sharply, though they don’t vanish: the model can still misread what it retrieved, so source links remain your friend.
Where you’ll meet this
Every time an assistant says “searching the web…” before answering: that’s RAG. Also: ChatGPT and Claude projects that answer from uploaded files, Copilot answering from your company’s SharePoint, Gemini Notebook (the tool Google called NotebookLM until July 2026) working over your sources, and customer-service bots that actually quote the real return policy.