What is retrieval-augmented generation (RAG)?
The plain-English definition, the 2020 paper that named it, and what it means for getting your business into AI answers.
Retrieval-augmented generation (RAG) means an AI looks things up before it answers. First it retrieves relevant text from a search index or document store; then it generates an answer using that text. It lets AI answer with fresher, more specific facts than it memorised in training, and it is how AI search engines can cite sources.
Where does the term RAG come from?
It comes from a 2020 paper by Patrick Lewis and colleagues, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks. NLP means natural language processing: getting computers to work with human language.
The paper's starting point: large language models store facts inside the model itself, but their ability to reach and use that knowledge precisely "is still limited." Its fix joined a model that writes text to a searchable index of Wikipedia, reached through a retriever. The authors report that this produced "more specific, diverse and factual language" than the same kind of model working from memory alone.
How does RAG work in AI search?
Think of an open-book exam instead of a closed-book one.
- You ask a question.
- Retrieve: the system searches an index for passages that match.
- Augment: it puts those passages in front of the model with your question.
- Generate: the model writes an answer from them, and can link where they came from.
The big engines describe the same pattern in their own words. Google defines grounding as "providing content from the Google Search index to the model at prompt time to improve factuality and relevancy." Google OpenAI runs OAI-SearchBot to surface sites in ChatGPT search, and ChatGPT-User to visit a page when a user's question calls for it. OpenAI
Why does RAG matter for my business?
Because the retrieve step decides who can be in the answer. A page that was never fetched, or can't be read, can't be retrieved — and a page that isn't retrieved can't be cited.
- Let the search bots in. See AI crawlers.
- Answer the question in a short, self-contained passage. In our experience, passages that make sense on their own are the ones that get lifted.
- Keep facts current on your own site. Retrieval works from indexed or freshly fetched pages, so a fix you publish can reach answers without waiting for a new model.
That last point is also why publishing the correct fact is the fix for an AI hallucination about your brand.
Sources
- Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (Lewis et al., 2020) — arXiv. Read Sep 24, 2026.
- Google's common crawlers — Google Search Central. Read Sep 24, 2026.
- Overview of OpenAI crawlers — OpenAI. Read Sep 24, 2026.
Common questions
Is RAG the same as training?
No. Training changes what the model has memorised and happens rarely. RAG hands the model fresh text at the moment it answers.
Does RAG stop AI from making mistakes?
It helps, but it doesn't guarantee accuracy. If the retrieved page is wrong or out of date, the answer can be wrong too.
Do ChatGPT and Google use RAG?
They describe the same idea. Google calls it grounding in the Google Search index, and OpenAI runs a search crawler and a user-triggered fetcher that supply pages to ChatGPT.
Related pages.
See what AI says about your business.
Enter your domain and the free AI visibility check shows where ChatGPT, Gemini, Perplexity and Google's AI answers mention you — and where they don't.