Definition
RAG combines a retrieval step (search a knowledge base or the live web for relevant documents) with a generation step (the LLM synthesises an answer using the retrieved context). This is what enables citations - the model can name the source it used. For SEO purposes, optimizing for RAG-based surfaces is closest to classic SEO plus extractability work.
Why it matters
RAG sits in the "Signals & ranking" layer of the AI search stack. Teams that handle it well get cited more, recommended more, and earn more of the AI-mediated revenue in their category. Teams that ignore it spend a year wondering why their content investment never moves the needle inside ChatGPT or Perplexity.
Related terms
- ChatGPT Search - OpenAI's live-retrieval mode inside ChatGPT - pulls real-time web results, cites them, and answers in the chat surface.
- Perplexity - A dedicated answer engine that retrieves the web in real time and synthesises answers with inline citations.
- AI Overviews - Google's generative-AI answer box at the top of search results, powered by Gemini. Now appears for more than half of qualifying US queries.
- Grounding - Augmenting an LLM's answer with retrieved evidence so the model can cite verifiable sources, rather than relying on training memory alone.
- Extractability - How easily an LLM can lift a clean, summarisable answer from your page. High extractability dramatically increases citation rate on grounded surfaces.
Apply it
The LLM SEO playbook ties every concept in this glossary into a single operating model. If you want to see how your brand performs across all the LLMs at once - mention rate, citation share, sentiment, rank - start with the free GEO audit or skip straight to a free Livesov account.