What is retrieval-augmented generation (RAG)?

It is the machinery behind grounded answers. The engine searches, selects passages, and generates an answer conditioned on them, which is why the sources it retrieves have more influence on the output than anything in the model's training data.

For anyone trying to be recommended, the consequence is that the fight is over retrieval. Being in the documents an engine pulls for a category question matters more than any property of your own writing.