Definition
Retrieval-augmented generation is a pattern where a system retrieves relevant, approved documents and passes them to an AI model as context, so answers are grounded in your own information.
RAG reduces invented answers by giving the model the facts it needs at the moment it needs them. Quality depends on the source content, the retrieval step and clear rules for what to do when the answer is not found.
Why it matters
Language models know a lot about the world but nothing about your policies, products or customers. Without that context they guess, and a confident guess is worse than no answer in most business settings.
RAG gives the model the right pages at the right moment and lets the application show where each answer came from. That makes answers checkable and keeps sensitive documents behind the same permissions they already have.
An example
An employee asks an internal assistant how many days of leave carry over to next year. The system searches the approved HR policy, retrieves the relevant clause, and the model answers in plain language with a link to the source paragraph. If the policy does not cover the question, the assistant says so and points to the HR contact.
Illustrative example, not a client project.
Common questions about Retrieval-augmented generation (RAG)
Does RAG stop AI from making things up?
It reduces it significantly but does not remove it. Good RAG systems also cite sources, refuse when nothing relevant is found and are tested against a set of known questions.
Is RAG better than fine-tuning a model?
They solve different problems. RAG supplies facts that change often; fine-tuning shapes style or behaviour. Most business assistants start with RAG because the knowledge stays editable.
Read more
Related terms