Retrieval-augmented generation
Also known as RAG
A method where a model fetches relevant documents at answer time and writes from them.
RAG is why a model can answer questions about events after its training cutoff, and why fresh, well-structured content can show up in answers quickly. The engine retrieves sources, then generates a response grounded in them.
It also explains the leverage AEO has. If you are among the documents retrieved for a query, you can influence the answer, even though you never see the index.