AI search · Advanced

Retrieval-augmented generation (RAG)

Retrieval-augmented generation (RAG) is a method where an AI system first retrieves relevant documents, then writes its answer using them. AI search features work this way, which is why being retrievable and quotable matters.

By Thryv StudioUpdated

Category
AI search
Level
Advanced
Also called
RAG
Steps
Retrieve, then generate
Why it's used
Current facts and citable sources

What is RAG?

RAG combines two systems. A retriever searches an index (the web, or a company's documents) for passages relevant to the question. A language model then writes an answer using those passages, often citing them. The approach was described in a 2020 research paper and is now the standard design for AI search.

Why it matters for SEO and AEO

If AI search is RAG, then being chosen is a two-step contest:

  1. Retrieval. Your page must be found as relevant. That depends on indexing, crawl access, relevance and authority, much like SEO.
  2. Generation. Your passage must be useful enough to use. That depends on clarity, specificity and structure.

Pages built from clear, self-contained sections do well at both steps, because each section can be retrieved and used on its own.

Best practices

  1. Give each section a descriptive heading and a direct first sentence.
  2. Avoid burying key facts in images, PDFs or scripts.
  3. Keep one topic per page, covered thoroughly.
  4. Use consistent terminology so retrieval matches your pages to the question.

Frequently asked questions

Is RAG the same as training?

No. Training changes the model itself. RAG leaves the model unchanged and supplies documents at the moment of answering.

Why do RAG answers still make mistakes?

Retrieval can return the wrong passages, or the model can misread them. Clear source content reduces both risks.

Browse the glossary

Chat on WhatsApp