A language model knows patterns from its training data, but it does not automatically know your product handbook, customer policies, or the decisions your team made last Tuesday. Retrieval-augmented generation (RAG) adds a search step before generation.
The basic loop
Documents are split into meaningful chunks and converted into vectors. When a user asks a question, the system finds the chunks with the closest meaning, places them in the model’s context, and asks for an answer grounded in those sources.
Why quality varies
Most failures happen before the model writes a word. Chunks may be too large, headings may be lost, or the search index may return plausible but irrelevant passages. Good RAG treats parsing, metadata, retrieval, citations, and evaluation as one product—not a single database feature.
The useful question is not “does it use RAG?” It is “can a reader verify why this answer was produced?”