Glossary
RAG
In short: Retrieval-augmented generation (RAG) is the architecture in which a system retrieves relevant documents and passes them to a language model to write a grounded answer.
Retrieval-augmented generation is the technical pattern behind most grounded AI answers. The flow has three stages: the user's question is turned into one or more search queries; a retriever fetches the most relevant passages from an index; and the language model is given those passages as context and asked to write an answer, usually with instructions to cite them.
The pattern matters for GEO because it changes which content wins. The retriever is typically a search engine or a vector index, so pages need to be indexable and semantically close to the question. The model then selects from the retrieved passages, so the useful part of a page needs to be self-contained and quotable: a paragraph that states the answer, the price or the location in plain terms, without depending on text elsewhere on the page.
RAG also explains some failure modes. If retrieval returns a directory listing with an old address, the model may repeat it confidently. If retrieval returns nothing about you, you are simply absent, however good your reputation offline.
Seoptist's GEO readiness audit checks for the conditions RAG depends on: crawler access, server-rendered content, FAQ blocks and question-shaped paragraphs, and consistency between your site and the fact card used to detect inaccurate answers.