Definition
RAG (retrieval-augmented generation) is a technique in which a language model first retrieves relevant documents — from a search index, a database or the live web — before generating an answer, and passes them to the model as context. The answer is then assembled not only from the model's memory but from current, verifiable sources the model can cite alongside the answer.
Almost every AI search experience works this way. ChatGPT search, Perplexity, Google AI Overviews, AI Mode and Copilot first run one or more searches for a question (query fan-out: one question is split into several sub-queries), retrieve the best matching pages or passages, and have the model write an answer from them with source references. The pages chosen in that retrieval step are the pages that get cited.
That has direct consequences for how you write content. Retrieval usually works at passage level: not the whole page, but individual paragraphs or sections are compared with the question, often through vector embeddings that compare meaning rather than literal words. A paragraph that answers a question on its own, naming the entity, scores higher than a paragraph that only makes sense in the context of the rest of the page.
The second step, generation, determines which retrieved passages actually end up in the answer. Models prefer sources that are concrete, consistent with other sources and recent, and that contain evidence (figures, examples, steps). Contradictory or vague passages are left out. To appear in AI answers, you therefore have to win both retrieval (findable and relevant) and selection (quotable and trustworthy).
RAG makes AI visibility influenceable in the short term: you do not have to wait for a model to be retrained, because a new or improved page can be retrieved the next day. Traze writes every page with independently quotable passages and a direct answer at the top, exactly the form that gets selected in the retrieval and generation steps of RAG systems.