Retrieval-Augmented Generation (RAG)
Also called: RAG
Retrieval-augmented generation is a technique where an AI system searches for relevant documents at query time and uses their contents to generate its answer, rather than relying only on training data.
RAG is why AEO is tractable at all. If answers came purely from training, influencing them would mean waiting for retraining. Because a retrieval step fetches live pages, changes to a website can affect answers within days.
Retrieval is also where most of the leverage sits. Whether your page is fetched, and whether the passage is extractable, are far more controllable than what a model absorbed months ago.
Why it matters
Understanding RAG explains both the opportunity and the timeline — and separates work that can move quickly from work that cannot move at all on a useful horizon.
An example
A user asks an assistant what a particular software plan costs this year. Instead of answering from training data that may be a year old, the assistant searches, fetches the vendor's current pricing page, and builds its answer from that page's contents. If the pricing only appears inside an interactive widget, the fetch returns little and the answer falls back on older information.
Common mistakes
- Assuming the model always searches. Many answers are produced without retrieval unless the question needs current information.
- Believing retrieval reads whatever a browser shows. It reads what the server returns.
Questions
- Where did the term RAG come from?
- It comes from a 2020 research paper that combined a retrieval step with a text generator so that answers could draw on documents fetched at query time. The approach now underpins most AI search products.
- Why does RAG matter for my website?
- Because it is the mechanism that lets your current pages influence answers now, rather than waiting for a model to be retrained. Being fetchable and readable at that retrieval step is where most practical leverage sits.