RAG · Retrieval-augmented generation
RAG (Retrieval-Augmented Generation) is an architecture that combines document search with a generative model: the system retrieves relevant passages, then generates an answer from them. It is the mechanism behind the citations in AI search engines.
Standalone definition · 36 words
How does RAG work?
Three steps. The query is converted into an embedding. The system retrieves the closest passages in an index. The model then generates the answer from those passages, which makes it possible to cite their source. The approach was described by Lewis et al. in 2020 for knowledge-intensive language tasks.
Why does RAG change the rules of SEO?
Content is no longer judged page by page but passage by passage, by semantic proximity to the question. An average page can provide the best passage. Writing quality at paragraph level becomes a visibility criterion.
How do you optimize for RAG?
Split content into self-contained sections, put the answer at the top of each section, name entities explicitly, add verifiable data, and keep pages crawlable and server-rendered. See chunking, answer-first and SSR.
Frequently asked questions
RAG vs fine-tuning: what is the difference?
Fine-tuning changes the model’s parameters. RAG adds information at query time without changing the model, which makes it possible to cite sources.
Do all AI engines use RAG?
Engines that cite web sources rely on retrieval before generation. Implementation details are not public.
Does RAG prevent errors?
It reduces them by providing sources, but the model can still misread them.
Related terms
Further reading
- Multimodal Geo Audit with Gemini-Embedding-2
- Bing AI Performance : how to exploit the data for ranker on ChatGPT, Copilot and other LLM
- Google Web Guide : The AI that rearranges the SERP
- ← Back to the GEO glossary (114 terms)