Glossary /AI search & GEO
Retrieval-Augmented Generation (RAG)
The technique behind cited AI answers: fetch relevant documents first, then have the model compose a response from them.
Retrieval-augmented generation is the architecture underneath every assistant that answers with citations. Rather than relying on what the model absorbed in training, the system searches a corpus for passages relevant to the question and gives them to the model as context, and the answer is composed from that retrieved material.
Knowing this explains what AI search rewards. Retrieval operates on chunks of documents, ranked by relevance to a sub-question, which is why a self-contained, well-labelled section is the unit that gets cited and why a page has to be crawlable and parseable to be in the corpus at all.
It also explains why a model can cite something published last week despite a training cutoff long before it. The freshness comes from retrieval, not from the model, so recently published pages compete on equal terms.