Retrieval-augmented generation (RAG)
Retrieval-augmented generation (RAG) is the technique AI systems use to ground answers in live sources: retrieve relevant documents first, then generate the answer from them, citing what was retrieved. It's why being retrievable — crawlable, structured, quotable — determines whether a brand gets cited.
In context
RAG explains a lot of GEO mechanics. When Perplexity or ChatGPT Search answers a shopping question, it doesn’t rely purely on training memory — it retrieves current pages, then composes from them. Three consequences for brands: your pages must be reachable by LLM crawlers at answer time, self-contained passages get quoted (which is why crisp definitions and FAQs work), and structured facts beat buried prose.
Training-data visibility still matters for what models “know” by default, but RAG is why fresh, retrievable content can win citations immediately rather than waiting for the next model release.
Related
See Generative Engine Optimization for the practice built on top of this mechanism.