A support assistant retrieves the current returns policy before answering a customer's question.
RAG adds a retrieval stage to a generation application. A search or retrieval system finds passages, records or other material; the application includes that information with the question. The language model generates an answer using the supplied context. IBM explains the architecture.
This differs from training material into model weights. Updating a document can change what retrieval supplies without retraining the base model. Retrieval and fine-tuning can also be used together.
Reliability still depends on the source and what gets retrieved. Ask to see the supporting passage, check its date and confirm that it supports the answer. An old policy or missing exception can produce a polished but unsuitable reply. For a small task, supplying the relevant text yourself may be enough; a document collection does not automatically require building a full RAG system.
Sources
Primary references checked 11 October 2026. This explanation is not a benchmark of a particular model or computer.