Retrieval-Augmented Generation (RAG)
TL;DR: What is Retrieval-Augmented Generation (RAG)?
Retrieval-augmented generation (RAG) is the architecture that lets an AI system fetch relevant documents at answer time and generate its response from them, rather than relying only on training data. RAG is the mechanism behind cited AI answers, and the reason AI visibility work pays off quickly.
Retrieval-Augmented Generation (RAG) explained
The approach was introduced in a 2020 research paper by Patrick Lewis and colleagues at Facebook AI Research, and it solved two chronic LLM problems at once: knowledge frozen at the training cutoff, and confident fabrication. Instead of answering purely from memory, a RAG system first retrieves relevant content, typically via semantic search over indexed passages, then generates its answer grounded in what it retrieved, often with citations pointing back to the sources.
Every major AI search experience runs on some version of this pattern. Perplexity, ChatGPT search, Google's AI Overviews and AI Mode, and Claude with search all retrieve before they write. The retrieval step is where visibility is contested: your content either gets pulled into the context the model writes from, or the answer gets composed entirely from other people's pages.
Two strategic consequences follow. Freshness works, because retrieval bypasses the training cutoff; a page published this week can be cited this week, with no wait for a model retrain. And passage quality beats page authority at the margin, because RAG systems retrieve chunks, not domains. A self-contained passage that cleanly answers the sub-question can beat a stronger site's meandering one, which is precisely the opening smaller brands should be exploiting.
In practice
RAG is the reason I can tell clients this work pays off fast. Your training-data footprint took years to form and moves slowly, but the retrieval door is open today: publish something clear, direct, and well-sourced, and an engine can cite it almost immediately. When a client's baseline shows near-zero citations, RAG is also the honest diagnosis. The systems are retrieving on every relevant prompt and repeatedly choosing other sources. That is fixable, and it is the exact gap the visibility work targets.
Common misconception
People often assume AI answers come entirely from what the model memorized in training. Actually, search-connected systems retrieve live content and build answers from it, which is why new pages can be cited within days.
Want the bigger picture? Start with the Founder’s Guide to SEO.