Retrieval-augmented generation

A language model on its own writes from what it learned in training, which is frozen at some date and blurry on specifics. Retrieval-augmented generation fixes that by putting fresh material in front of the model before it writes: run a search, fetch the top pages, place their text in the prompt, and ask for an answer grounded in them, often with instructions to cite. Every assistant that shows sources works this way. The pattern explains most of what a store can do about AI visibility. The retrieval step is a search, so the same signals that put a page on a results page put it in the model's context. The reading step favours pages whose relevant passage is easy to find, which is why a direct answer near the top matters. The writing step quotes and paraphrases, which is why specific, checkable facts survive and vague marketing does not. A page that cannot be fetched by the engine's crawler never enters the loop at all, whatever its quality. Before changing store access, follow the crawler-specific Shopify guidance rather than assuming every retrieval path uses the same bot.