Tornado SEO

Retrieval-augmented generation (RAG)

ThibaultBy Thibault · Updated · 2 min read

Definition

**Retrieval-augmented generation (RAG)** is a technique where an AI model first retrieves relevant documents from an external source, then writes its answer from that content instead of relying only on what it learned in training.

On this page
  1. Why does RAG matter for SEO?
  2. How does retrieval-augmented generation work?
  3. What should you do about RAG?

Why does RAG matter for SEO?

RAG is the mechanism that lets an AI cite your page. A model's training data is frozen at a date, so when ChatGPT searches the web or Google builds an AI Overview, the model fetches pages, reads passages and writes its answer from them. Retrieval is the door through which fresh content gets into an AI answer. If your page is not retrieved, it cannot be quoted, however good it is. This is why I see GEO as an extension of SEO: the retrieval step still depends on being crawled, indexed and judged relevant.

How does retrieval-augmented generation work?

  1. The question is turned into one or more search queries. Google calls this query fan-out: AI Overviews and AI Mode can issue multiple related searches across subtopics and data sources.
  2. A retriever pulls candidate pages or passages from an index.
  3. The most relevant passages are added to the context the model reads.
  4. The model writes the answer from those passages and can link the sources it used.

The term comes from a paper first submitted in May 2020 by Patrick Lewis and 11 co-authors, accepted at NeurIPS 2020. Their system paired a language model with a dense index of Wikipedia, and they reported answers that were more specific and more factual than a model working from memory alone. Take a simple case: you ask an AI how long a core update takes. Instead of guessing, it runs a few searches, reads the top passages and reuses the sentence that answers in one line. The model picks passages, not whole pages, so the paragraph that stands on its own wins.

What should you do about RAG?

Thibault's take I treat RAG as a reading test, and accessibility comes before visibility. For Google, a page must be indexed and eligible to be shown with a snippet to appear as a supporting link in AI features, with no extra technical requirement. So I start with the basics: is the text in the HTML, or does it only show up after JavaScript runs? If the crawler does not see the text, there is nothing to retrieve. Then I rewrite each section so its first sentence answers alone, with a clear question as the heading. That is what gets lifted into an answer. There is no absolute here, but I have never seen a page cited that the robot could not read.

Sources

  1. arxiv.org/abs/2005.11401
  2. developers.google.com/search/docs/appearance/ai-features

Frequently asked questions

Is RAG the same thing as AI search?

RAG is the technique behind most AI search features. ChatGPT search, AI Overviews and AI Mode all retrieve web content before writing an answer, which is a form of retrieval-augmented generation.

Does RAG replace traditional SEO?

No, it depends on it. The retrieval step relies on pages being crawlable, indexed and relevant, which are classic SEO work. What changes is the format: passages that answer alone get reused more easily.

Does RAG stop AI models from making mistakes?

It reduces errors but does not remove them. The model can still misread a passage or retrieve a weak source, which is why clear, precise wording on your page helps it quote you correctly.

Thibault

WRITTEN BY

Thibault Besson-Magdelain, founder of Sorank, SEO and GEO expert

I'm Thibault Besson-Magdelain, founder of Sorank, SEO and GEO expert. I help companies be visible, understood and cited by AI, through SEO, data and semantic structuring.

FREE, EVERY MORNING

Stay ahead of every search update

Google, ChatGPT, Gemini, Claude: what changed, why it matters, and what to do about it. Every morning, in 3 minutes, free.