What is retrieval-augmented generation (RAG)?
Also called: RAG
Definition
Retrieval-augmented generation (RAG) is a technique in which an AI system first retrieves documents relevant to a question, then passes them to a language model to write the answer. It lets the model use information it was never trained on and point to its sources.
Explained
The term comes from a 2020 research paper by Patrick Lewis and colleagues at Facebook AI Research. AI search products are large-scale RAG systems: a search step finds candidate pages, a model reads the most relevant passages, and the answer is written from them.
Two consequences matter for GEO. You must be retrievable, meaning indexed and ranking for the underlying searches. And the passage that gets retrieved must say something useful on its own, because the model often sees a chunk of your page, not the whole thing.
Example
A buyer asks Perplexity for "the cheapest SOC 2 compliance tool for startups". It searches, picks a handful of pages, extracts the passages that mention pricing and startups, and writes the answer from those passages, citing each one.