RAG stands for retrieval-augmented generation. It sounds technical, but the idea behind it is simple: instead of asking an AI to answer purely from what it memorized during training, you hand it the relevant page from your own documentation first, then ask it to answer using that.
The two halves of the name
Retrieval is the "look this up" step. When a visitor asks a question, the system searches your connected docs and pulls out the passages most likely to contain the answer — not the whole document, just the relevant piece.
Generation is the "write a reply" step. The AI takes those retrieved passages and the visitor's question, and writes a natural-language answer grounded in what it was just handed.
Put together: retrieve the right information, then generate an answer from it.
Why this matters more than which AI model you use
A generic AI chatbot answers from general training data — it doesn't know your pricing page changed last week, or that your product has a feature that didn't exist when it was trained. A RAG-based chatbot answers from whatever you've connected right now, so it's current by construction, not by luck.
It also means an answer can be traced back to a real source. If the chatbot says something wrong, you can go find the document it pulled from and see exactly why — instead of trying to debug a black box.
What this looks like in Lumen Chat
When you connect Notion, Google Drive, or your website, that content becomes the retrieval side of the pipeline. Every answer the chatbot gives is generated from passages retrieved from your own docs, not from the AI's general knowledge — which is also why it says "I don't know" when your docs don't cover something, instead of making an answer up. See our post on grounded vs. hallucinated answers for more on that distinction.
You don't need to understand the mechanics to use a RAG-based chatbot well — but knowing that "answers come from what you connected" explains most of what you'll see day to day: better docs mean better answers, and gaps in your docs show up as gaps in the chatbot's answers.
Related reading
- Chunking and embeddings 101 — a closer look at how your content gets indexed for retrieval.
- How vector search finds the right answer — the retrieval half of RAG in more depth.
- Turning your docs into an AI chatbot — putting this into practice on your own content.