RAG

RAG stands for Retrieval-Augmented Generation. It is a technique in which a language model doesn't only draw on what it learned during training, but first looks up current or specific information and then uses it to formulate an answer. So the model "retrieves" and "generates" in a single process.

That may sound like a technical detail, but it is exactly why AI search environments such as Perplexity, ChatGPT Search and Google AI Overviews can refer to websites. Without RAG, an LLM could only answer based on its training data, which quickly goes out of date and doesn't cite sources.

How does RAG work, step by step?

The exact workings differ per system, but the general process looks roughly like this:

1. Analysing the question. The system determines which information is needed and turns the question into one or more searches.
2. Retrieving information. It searches an index, for example the web, an internal document library or a product catalogue, and selects relevant passages.
3. Providing context. Those passages are passed to the language model together with the original question.
4. Formulating the answer. The model writes an answer based on that context and often shows the sources used as a citation.

Why RAG matters for your website

With RAG, the retrieval step decides which sources the model gets to see. If your page doesn't come up in that step, it can't end up in the answer either. That means the classic basics still apply: your page needs to be crawlable, indexed and recognised as relevant.

The form of your content matters too. RAG systems usually work with fragments rather than full pages. A paragraph that answers a question clearly and on its own is easier to reuse than an answer spread across a long text. A clear structure with good subheadings helps, as does structured data that makes facts such as opening hours or prices unambiguous.

An example

An accountancy firm in Manchester publishes a clear page on the VAT rules for small businesses, with a short answer for each situation. When someone asks an AI assistant whether, as a new business, they need to register for VAT, the system can select exactly that passage in the retrieval step and cite the firm as a source. A general "Our services" page wouldn't answer that question and so wouldn't be retrieved.

RAG isn't only used by search engines, by the way. Companies also use it for internal chatbots that answer based on their own manuals, product sheets or customer files. The principle is the same: the better structured and more up to date the sources, the more reliable the answer.

RAG and GEO

The growing field of GEO is largely about understanding RAG: how do you make sure your content is found, selected and represented correctly? There is no fixed formula, and systems change constantly. What does seem consistent is that clear, reliable and easy-to-find information gives you the best starting position. Your authority as a source probably also plays a part in which pages get selected.

Want to see how to prepare your website for this new way of searching? Take a look at how the Spacebird platform for SEO and AI visibility brings both worlds together.