RAG (Retrieval-Augmented Generation)
The model producing its answer not only from memory but from documents it retrieves at that moment (web search, a database). Perplexity works this way entirely, ChatGPT and Gemini often do; that's why your page needs to be quotable.
RAG stands for retrieval-augmented generation. When the model receives a question, it first searches for and fetches relevant documents (retrieval), then writes the answer based on those documents (generation). Training data may be frozen years in the past; with RAG, the answer rests on current sources.
What it means for a business
In an answer that uses web search, the model reads a handful of pages from the search results and builds the answer from them. If your page is one of that handful, you appear in the answer and are usually shown as a source. If it isn't, the model may leave you out of the answer even if it remembers you from training data.
This requires two things:
- Being found: your page must be in the search index (Google, Bing) and match the question.
- Being quotable: when the page opens, the answer to the question must be in the first paragraph, clear and short. The model won't hunt for a sentence in the middle of a long page.
Which engines
Perplexity uses RAG on every question and lists its sources. ChatGPT and Gemini turn to search on current or local questions and mostly don't on general ones. Claude can search but is more cautious.
The practical consequence
The spread of RAG is bringing classic SEO and GEO closer together: a page that does well in the search index also gets into AI answers. The difference is in the form of the page: an answer rather than a list, a description rather than a slogan.
Related term: Citation