Retrieval-Augmented Generation (RAG)
RAG is a technique where an AI model actively retrieves external sources at answer time so responses are accurate and verifiable.
Retrieval-Augmented Generation — RAG — is the architecture most modern AI search engines use under the hood. Instead of the LLM answering from memory, the system first searches a knowledge base (the web, a product catalog, a document archive), selects relevant passages, and feeds those as context to the language model. The model builds its answer around those real, current facts. For GEO, RAG is the main trigger that gets your page cited. Perplexity, ChatGPT Search, Google AI Overview and Google AI Mode are all variants of RAG.
Example
A user asks Perplexity: 'What did an Antminer S21 cost in March 2026?'. Perplexity performs an active web search (retrieval), finds antminerdistribution.com among others, extracts the pricing information, and generates a centralised answer with that source as citation. Without RAG the LLM would never have known; with RAG the page is explicitly named.
Frequently asked questions
What's the difference between RAG and plain web search?
Plain web search gives you a list of links; you read yourself. With RAG, an LLM reads the found pages and turns them into one flowing answer with citations. For content owners: in RAG environments you influence not just ranking but whether you're literally quoted.
How do I optimise for RAG retrieval?
Make your content easy to find and easy to parse for crawlers. Strong classic SEO (crawlability, authority, freshness), clear h2/h3 structure, TL;DR blocks, schema.org markup, and answers that literally rephrase the user's question.
Related terms
Further reading
- → Our service: GEO
- → Blog: What is GEO? Generative Engine Optimization explained