RAG(retrieval-augmented generation)
Also called: retrieval augmented generation
Definition
RAG, short for Retrieval-Augmented Generation, is a technique that combines two steps: retrieving relevant documents, web pages for example, then generating an answer with a language model from those documents. The method was formalized in 2020 by researchers at Facebook AI.
It's the principle behind ChatGPT Search, Perplexity or AI Overviews: the engine retrieves pages, extracts the useful passages, then writes. For a brand, this changes a lot: a page can be read and picked up as soon as it's indexed, without waiting for a new model to be trained.
Example
You ask “what are Qonto's prices?” The engine retrieves the pricing page and two comparisons, then writes an answer that repeats the numbers it just read.
What it means for your AI visibility
With RAG, part of your AI visibility is decided today, and not just in training data. A clear page, readable without JavaScript and open to crawlers, can be cited quickly.
Related terms
- GroundingGrounding means basing an AI's answer on verifiable sources rather than on its memory alone.
- LLMAn LLM is an AI model trained on huge volumes of text to understand and produce language. It is the engine behind ChatGPT, Claude or Gemini.
- Query fan-outQuery fan-out is the technique a generative engine uses to split a question into several searches run in parallel.
- Citable contentCitable content is written so AI can reuse it as is: one idea per paragraph, precise facts, a direct answer to a question.
- AI crawlerAn AI crawler is a bot that browses the web on behalf of an AI company, to train its models or feed its answers.
Updated on · All glossary terms