FYI, RAG helps AI answer from a source.
Retrieval-augmented generation finds relevant information and supplies it to a model before an answer is generated. It can ground an answer in selected sources, but retrieval mistakes and unsupported claims are still possible.
Retrieve relevant material, then generate with it. A source marker helps locate evidence; it does not guarantee that the source is current or that the answer is correct.
View diagramSee it happen
Give the answer something to refer to.
Your question“How long do I have to return an item?”
I don’t have this shop’s policy. Check its current return terms.
Retrieval supplies relevant material before the model answers. It helps ground an answer; it does not guarantee the source is current or the answer is correct.
Scripted illustration, not a live AI result. Example details are fictional.
Give the answer somewhere to come from
In a fictional restaurant guide, you ask: “Which places have a quiet room?” A retrieval step searches the guide and selects relevant passages. Those passages accompany your question when the model generates its answer.
That is the basic RAG pattern: retrieve relevant information, then generate with it. The source might be a document collection, a database, or a search result. It does not have to be the public internet.
This supplies information at answer time. It is not the same as retraining the model on the entire guide.
A common mix-up
RAG does not guarantee truth. The system might retrieve the wrong restaurant, miss an important sentence, use an outdated guide, or produce a claim that the selected passage does not support.
A citation helps you inspect the evidence. It is not a certificate that the answer matches it. “Has a private room” also does not automatically mean “is quiet” or “has a table available tonight.”
Why it matters
Check three things: the source, the selected passage, and the answer. Is the document current? Is the passage about the right place? Does it actually support the conclusion?
If the guide never mentions room noise, a useful response should make that gap clear. Getting more documents is not necessarily the fix; finding the right information and preserving its meaning matters more.