What is RAG in AI?
The short answer
RAG stands for retrieval-augmented generation. It means the AI looks things up in your documents before it answers, instead of relying only on what it memorised during training. Think of it as giving the model an open book rather than asking it to recall everything from memory.
How RAG actually works, in three steps
- 1
Your documents are indexed
Policies, tickets, contracts, wiki pages — chopped into passages and stored so they can be searched by meaning, not just keywords.
- 2
The question triggers a search
Someone asks "what's our refund window for enterprise?" The system retrieves the handful of passages most likely to answer it.
- 3
The model answers from those passages
Those passages are handed to the model along with the question. It writes the answer using them — and can point back at the source.
RAG vs fine-tuning vs prompting
These three get muddled constantly in vendor decks. They solve different problems.
| Approach | What it changes | Use it when |
|---|---|---|
| RAG | What the model can see when answering | Answers must come from your own, changing content |
| Fine-tuning | How the model behaves, its style and format | You need a consistent tone or output structure |
| Prompting | The instructions given for one task | The knowledge is small enough to paste in |
Where RAG quietly goes wrong
- Stale sources. The index only knows what it last ingested. An old policy document produces a confident, cited, wrong answer.
- Bad retrieval, good grammar. If the search step returns the wrong passage, the answer still reads beautifully.
- Permissions leak. If the index ignores who is allowed to see what, RAG becomes a very fast way to surface documents people shouldn't read.
Say this in the meeting
“So RAG means answers are grounded in our own documents rather than the model's memory. What happens when retrieval finds nothing relevant — does it say so, or does it guess?”
Frequently asked questions
What does RAG stand for?
RAG stands for retrieval-augmented generation. The model retrieves relevant documents from your own content first, then generates its answer using them.
What is RAG in simple terms?
It's an open-book exam for an AI. Instead of answering from memory, the system looks up your documents, pulls the relevant passages, and writes the answer from those passages.
Why do vendors talk about RAG so much?
Because it is the cheapest credible way to make a general-purpose model answer questions about your private data — no retraining, and answers can cite a source.
Is RAG the same as fine-tuning?
No. Fine-tuning changes the model's behaviour and style by training it further. RAG changes what information the model can see at the moment it answers. Most business use cases want RAG first.
Does RAG stop AI hallucinations?
It reduces them, but does not eliminate them. If retrieval returns the wrong passage, or your source documents are out of date, the answer is confidently wrong with a citation attached.
What should I ask a vendor selling RAG?
Ask what happens when retrieval finds nothing, whether every answer links to its source document, how often the index refreshes, and who can see which documents.
RAG is one term. There are about forty more.
Falquin teaches the AI vocabulary that shows up in business meetings — five minutes a day, every term with a plain-English meaning and a line you can actually say out loud.
Got a meeting where RAG is on the agenda? Paste your agenda into Meeting Prep and get every term you'll hear.