Retrieval-Augmented Generation (RAG)

RAG is the architecture pattern where an AI engine fetches live web sources, then synthesises an answer from them.

Retrieval-Augmented Generation (RAG) is the dominant architecture pattern for AI search engines: a user query is rewritten into one or more search queries, top-N results are fetched, those documents are split into chunks, the chunks are embedded and ranked for relevance, and the most relevant chunks are passed as context to a generative model that produces the final answer. Perplexity, Google AI Overviews and ChatGPT-with-browsing are all RAG-based. RAG is what makes citation-worthy content matter — the model is literally reading and quoting your page in real time.

Related terms