Retrieval-augmented generation (RAG)
Short definition
RAG lets an AI model pull information from a trusted set of documents before it writes an answer, not from memory alone.
Retrieval-augmented generation, RAG, has a language model look up a specific document or database at the moment of the query rather than relying only on what it learned during training. It searches the relevant source first, then writes an answer grounded in it.
This reduces the model's tendency to make things up, hallucinate, on a topic it does not know or knows outdated, because the answer is now drawn from a shown source rather than general patterns.
A business can connect its own product catalogue, FAQ text or policy documents to a RAG system, so a customer assistant speaks only from the business's own accurate data.
RAG is often confused with fine-tuning, but fine-tuning retrains the model's own weights whilst RAG leaves the model untouched and simply shows it a relevant document at query time. In practice, you can tell an assistant uses RAG when its answers include a source reference or a page citation; without one, it is likely relying only on its original training.
Why it matters
Setting up RAG guarantees an AI assistant talks from the business's own accurate data rather than general knowledge from the internet. It sharply cuts the risk of quoting a wrong price or the wrong policy.
Illustrative example
A virtual POS provider connected its support assistant to its own technical documentation through RAG, and it stopped suggesting outdated integration steps.
