RAG (Retrieval-Augmented Generation)
A technique that allows LLMs to pull in real-time information from external sources (like a website) before generating an answer.
What is RAG?
Retrieval-augmented generation is an architecture in which a language model is given relevant documents at answer time instead of relying only on what it memorised during training. The system first turns the user's request into one or more search queries, retrieves matching content from an index or the live web, selects and trims the most relevant passages, and then passes those passages to the model along with the question. The model writes its answer grounded in that material and, in answer engines, attaches citations to the passages it used.
RAG is what makes AI search possible. Without it, a model can only describe the world as it was at its training cutoff. With it, ChatGPT search, Perplexity, Google AI Overviews, Copilot and Gemini can answer questions about events, products and prices that post-date the model. It is also why crawler access matters: the retrieval step is performed by agents such as OAI-SearchBot, ChatGPT-User, PerplexityBot and ClaudeBot, and a page those agents cannot fetch cannot be retrieved or cited.
RAG does not guarantee accuracy. The model can still misread a passage or blend sources, and retrieval quality depends on how clearly the page states its answer. Pages that put the answer near the top, in plain language, with specific and verifiable detail, are easier to retrieve and easier to quote.
Why it matters for AI search
Almost every GEO recommendation traces back to how RAG works. Allowing retrieval crawlers, rendering content server-side, writing answer-first paragraphs, adding structured data and keeping facts current all exist to make a page easy to retrieve and easy to quote at answer time. Understanding RAG also clarifies the two paths to visibility: being in the training data (slow, cumulative) and being retrieved live (fast, controllable), and why the second is where most day-to-day GEO work happens.
Related terms
LLM (Large Language Model)
A type of AI trained on vast amounts of text data to understand and generate human-like language (e.g., GPT-4, Claude 3).
Synthetic Traffic
Web traffic where the primary visitor is an AI agent or crawler acting on behalf of a user.
Source Attestation
The process by which an AI model verifies the authority and truthfulness of a source before citing it in an answer.
Hallucination
When an AI model generates factually incorrect information confidently. Monitoring hallucinations is critical for brand safety.
Frequently asked questions
Which AI search tools use RAG?+
ChatGPT with search enabled, Perplexity, Google AI Overviews and AI Mode, Gemini, Microsoft Copilot and Claude with web search all retrieve live content before answering. Plain chat modes without browsing rely on training data alone.
How do I make my pages RAG-friendly?+
Allow AI retrieval crawlers in robots.txt, serve content in the initial HTML rather than only after JavaScript runs, open each page with a direct answer, use descriptive headings, keep facts specific and current, and add relevant structured data.