Definition: Retrieval augmentation is the technique of retrieving relevant information from external sources and adding it to a model's prompt so the answer is grounded in that information. It is the core of retrieval-augmented generation (RAG).Why It Matters: Models have training cutoffs and no knowledge of private data. Retrieval lets them answer from current, internal sources and provide citations, which reduces made-up answers. Quality depends on retrieval: if the wrong passages are found, the answer will be wrong.Key Characteristics: A RAG system has an ingestion pipeline, an index, a retriever, a prompt that includes the retrieved text, and a generator. Enhancements include re-ranking, query rewriting, hybrid search and permission filtering.
Content is chunked, embedded and indexed ahead of time. When a user asks a question, the system may rewrite it and searches the index for relevant passages. Results are filtered by permissions and re-ranked. The top passages are inserted into the prompt with instructions to answer only from them and cite sources. The model generates the answer, and checks verify that claims are supported. Logs capture the question, sources and answer for evaluation. Teams measure retrieval quality and answer faithfulness separately, to find where problems originate.ProsAnswers are current, specific to the organization and traceable to sources.ConsPoor retrieval, stale indexes or weak prompts still produce wrong answers.
Internal Assistant: Employees ask about policies and receive cited answers from current documents.Customer Support: An assistant answers from the latest product documentation.Research Review: Analysts query a library of reports and see the passages used.
Open-Domain Question Answering (2017-2019): Research combined retrievers with readers.RAG Paper (2020): The term retrieval-augmented generation was introduced in research.Enterprise Adoption (2022-2023): Vector databases and LLM APIs made RAG a common enterprise pattern.Current Practice and Trends (2023-Present): Systems use hybrid search, re-ranking and agentic retrieval, with evaluation and permissions.
When to Use: Use retrieval augmentation when answers must reflect current or private information with citations.Designing for Reliability: Evaluate retrieval and generation separately, and handle empty results gracefully.Operating at Scale: Keep indexes fresh, monitor quality and cache frequent queries.Governance and Risk: Enforce source permissions, log sources used and review content approved for retrieval.
