What is retrieval augmented generation (RAG)?

A plain language model answers from its training data, which can be out of date or invented. RAG adds a retrieval step first: the system searches a source, pulls back the most relevant passages, and gives the model those passages to write from.

This is why answer engines can cite specific pages. When Perplexity or ChatGPT search answers a question, it retrieves candidate pages, then generates an answer grounded in what it pulled back, with links to the sources.

For a business, RAG is the reason being indexable and quotable matters. If your page is not in the source the engine retrieves from, it cannot be pulled into the answer, no matter how good the content is.