What is Retrieval-Augmented Generation?
RAG is the technique where a language model retrieves relevant documents or passages at query time and grounds its answer in them, instead of relying only on facts it memorized during training — it's why being retrievable is a prerequisite for being cited.
A large language model's training data is frozen at a point in time and can't reliably recall specific, current facts about every brand on the web. Retrieval-Augmented Generation solves this by adding a live retrieval step: before (or while) generating an answer, the system searches an index — the live web, a vector database, a proprietary corpus — pulls back the most relevant passages, and feeds them to the model as context alongside the user's question.
Perplexity, Google AI Overviews, ChatGPT's search mode, and most "answer with citations" products are RAG systems in some form. This is the mechanical reason citations exist at all: the model is quoting or paraphrasing a specific retrieved document, and typically links to it, which is precisely the moment a business becomes visible (or invisible) in an AI answer.
For GEO, RAG has a direct, practical implication: being retrievable is the prerequisite for being cited. A page has to be crawlable, indexed by whatever retrieval system the engine uses, and structured so the retrieved passage reads as a clear, self-contained answer. A brand with excellent content that a retrieval system never surfaces is functionally invisible to that engine, no matter how good the writing is.