Retrieval-Augmented Generation (RAG)
Retrieving source material before generating an answer.
Technical Term
Full explanation
Retrieval-Augmented Generation, or RAG, is a technique where an AI system retrieves relevant information from an external source before generating a response, so the answer is grounded in current information rather than training data alone.
RAG works in two stages. First, the system retrieves a set of documents or passages likely to be relevant to the question. Second, it generates a response using both that retrieved material and its own trained knowledge. The retrieval stage determines what evidence is even available to the generation stage.
This is the mechanism underneath Grounding: it is one specific way a system connects an answer to external information rather than relying purely on patterns learned during training. Most consumer AI search and chat products now use some form of RAG for time-sensitive or specific queries.
Why this matters
It is the reason publishing still works. If systems only used training data, nothing published today could affect an answer for years.
Example
Asked about a company's current pricing, a RAG-based system retrieves the live pricing page rather than answering from training data that may be a year out of date, and cites it in the response.
Where this sits
Retrieval-Augmented Generation (RAG) belongs to Retrieval, Evidence & Consensus, one of the seven layers of the Aingworth Glossary. Information exists, then it is retrieved, then sources are selected, then they corroborate one another, then consensus emerges. This is where Aingworth's Consensus Over Rankings thesis lives.
Continue with: The Stakes Test: Why Credibility Decides High-Value Deals.
Related terms
Ready to see where your brand stands?
Start with the free diagnostic to see which condition of the Recognition Triangle is leaking, or read the frameworks these terms belong to.