Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation (RAG) is a method where an AI system first retrieves relevant documents, such as web pages, and then uses a large language model to write an answer based on them. It lets AI tools give more current, specific answers and cite their sources, rather than relying only on training data.

Written by Paul Hoda, SEO consultant since 2005. Last updated 8 November 2026.

How RAG works

A RAG system has two main parts: a retriever and a generator. A simplified version looks like this:

  1. Understand the question: the system interprets the prompt and may split it into several searches, a process known as query fan-out.
  2. Retrieve: it searches an index, such as a web search index or a company’s documents, for relevant passages. Many systems use vector search, which matches meaning rather than exact words.
  3. Select: it ranks the passages and picks the most useful ones.
  4. Generate: the large language model reads those passages and writes an answer, often with links to the sources.

RAG is not only used in public AI search. Businesses use it to build chatbots that answer questions from their own help centre, policy documents or product data. The same principle applies: the answer is only as good as the documents the system can find and the passages it selects.

Why AI search uses RAG

LLMs on their own have a training cut-off and can invent facts. Retrieval gives them up-to-date information to work from, which reduces, but does not remove, the risk of errors. Tools such as Google’s AI Overviews and AI Mode, ChatGPT search and Perplexity all combine search-style retrieval with generation, although each works differently and the details are not fully public.

Why RAG matters for SEO

In a RAG system, your page has to pass two tests: it must be retrieved, then it must be useful enough to be used and cited. This has practical consequences:

  • Classic SEO still counts: if a page is not crawled and indexed, it cannot be retrieved. Many systems lean on existing search indexes.
  • Passages matter: retrievers often work with chunks of a page, not the whole thing. Clear sections that each answer one question are easier to retrieve, which is the idea behind content chunking.
  • Clarity wins: direct, factual statements are easier for a model to quote than vague copy.
  • Access matters: some AI tools use their own crawlers, so check your robots.txt rules for AI crawlers.

How to make content RAG-friendly

  • Use descriptive headings that match real questions.
  • Answer the question in the first sentence or two of each section.
  • Keep key facts in HTML text, not only in images or scripts.
  • Include specifics such as figures, definitions and steps that add information gain.
  • Keep pages accurate and updated, since retrieved content can be quoted directly.

Limits

RAG does not guarantee accuracy. A model can still misread or misquote a source, or blend several sources wrongly. Site owners cannot control which passages are chosen, only make their content easier to find, understand and trust.

Key Takeaways

  • RAG retrieves relevant documents first, then generates an answer from them.
  • It helps AI tools give current answers and cite sources.
  • Pages must be crawlable and indexed before they can be retrieved.
  • Clear, self-contained sections are easier to retrieve and quote.
  • RAG reduces but does not eliminate AI errors.

Related Terms

Want to know where your site stands? Book a free SEO health check, or keep learning in the SEO Encyclopedia.