Retrieval-Augmented Generation (RAG)

Retrieval-augmented generation, usually shortened to RAG, is the common arrangement in which an AI system searches for relevant pages, reads them, and composes its answer from what it has just found — rather than relying only on what it absorbed during training. When an assistant answers a question about a local business, this is generally what is happening under the surface.

The distinction matters considerably more than the acronym does. Anything a model learned during training is fixed: it reflects the web as it stood when the model was built, and you cannot change it after the fact. Anything retrieved at the moment of the question is live. It is read fresh, from your site as it exists today.

That is the whole reason fixing your website works at all. If these systems relied purely on training, a small business would have no lever: the picture would be set, and there would be nothing to do but wait years for the next model. Because most assistants retrieve, the page you correct this afternoon can be the page that gets read into an answer next month.

It also explains why access matters so much. Retrieval means a machine has to be able to reach and read the page at the moment it is needed. A crawler turned away, a page that arrives empty, text locked inside a PDF — each of these removes you from the pool that can be retrieved, whatever is written on the page itself.

The human translation is the difference between somebody recommending you from memory and somebody looking you up before they answer. The first you cannot influence. The second you can, by making sure that what they find is clear, current, and worth repeating.