RAG stands for retrieval-augmented generation. A system finds material relevant to a question and supplies it to a language model as context for generating an answer. It brings your records into the answering process without requiring the model to have learned those records during its original training.
The result depends on several things working together: suitable sources, useful retrieval and an answer that faithfully represents the material found. RAG does not make an outdated policy current or turn an unsupported statement into evidence.
A small-business example
A workshop wants staff to ask questions about its internal procedures. A RAG system searches the approved procedure notes and returns an answer with links to the passages used.
For a straightforward query, this can make a long manual easier to navigate. For a query spanning an exception and a recent change, staff must be able to open the sources and check the interpretation. If the collection contains no applicable procedure, the useful response is to identify that gap.
Try this
Before building an answering tool, assemble ten realistic questions with the source passages that should support their answers. Include conflicting versions, an exception and an unanswerable question. Use this small set to examine both what the system retrieves and what it says about the retrieved material.
Retrieval finds the evidence. Source of truth establishes which material is authoritative, and Human review checks the answer's claims against it.