RAG combines search and text generation. The method runs in two steps: first, a retriever searches an index for relevant passages. Then a language model writes the answer based on those passages.
The term was coined by a research team led by Patrick Lewis in May 2020 in the paper “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks”. Their model drew on an index of Wikipedia passages and produced more specific and factually accurate answers.
AI search works on a similar principle. For AI Overviews and AI Mode, Google describes how the system retrieves pages from its index and generates an answer with links. For website owners this means:
Content must be crawlable and indexed, otherwise no retriever can find it.
Clearly structured, self-contained paragraphs are easier to retrieve as passages.
Our article on AI Overviews and AI Mode shows how this works at Google.