What is retrieval-augmented generation?
RAG retrieves external information and supplies it to a model as context for generating an answer.
A customer-support system might first look up the company's returns policy, then answer whether an opened item can be returned. The retrieved clauses give the model material beyond its previously learned patterns.
Retrieval does not have to mean live web search
The information can come from internal documents or an external database. A RAG system need not search the public web on every request. A search API is one possible source of retrieved material, rather than the definition of the whole workflow.
Why can a RAG answer still be wrong?
The system may retrieve the wrong document, use an old policy or miss a condition. The model can also misread a relevant passage.
Check both stages: does the retrieved material apply to the question, and does it support the answer as written? Retain the source date and cited passage. Retrieval supplies evidence to examine; it does not complete that examination automatically.
Sources and scope
Source-language entry updated 2026-09-20; no separate source-check date is recorded. This adaptation preserves that provenance and does not represent a new audit of the cited pages.