What is Retrieval-Augmented Generation (RAG)?
Retrieval-Augmented Generation (RAG) is a method in which a language model retrieves relevant text passages from a document collection or the web before writing an answer and uses them as its basis. This lets the model answer with information that is not in its training data.
The term goes back to a research paper by Lewis and colleagues, published at NeurIPS 2020. The basic pattern has since become standard: a request is turned into a search, the best results are split into chunks, and the most relevant chunks go to the model together with the question.
For websites less follows from this than is often claimed. The model rarely reads a whole page, it reads individual passages. That self-contained paragraphs are therefore used more often is not proven. Google writes in its own guide to AI features that there is no requirement to break content into tiny pieces for AI; its systems find the relevant part even on pages covering several topics.
AI search products such as ChatGPT with web search or Perplexity work on this principle, but vendors do not publish their exact implementation. Claims about the ideal passage length are therefore guesses.
What it means for your website
For your own page this mainly means: the content has to be in the delivered HTML, since many retrieval systems do not execute JavaScript. Well-structured paragraphs, tables and lists help readers; that they are easier for AI search to cite is not proven.
Related terms
Sources
Terms help you understand. Whether AI crawlers can reach your site is something you measure.
Check your website for freefree · no signup · no credit card