In one sentence
RAG means retrieve a few relevant documents, feed them to a language model, and have the model write an answer grounded in those documents.
Why it matters for AI visibility
Most modern AI search engines use retrieval in some form. The specifics differ.
- Perplexity is a retrieval-first product: every answer is grounded in cited sources.
- Google AI Overview and Google AI Mode retrieve from Google's index.
- Microsoft Copilot grounds against Bing.
- ChatGPT, Google Gemini, and Claude can use web-search tools that make them behave like RAG systems at runtime.
Not all AI engines use RAG the same way. Some retrieve before every answer. Some retrieve only when the model decides a web search is needed. Some pass raw HTML to the model, others pass pre-processed summaries.
Two practical implications follow. On-page content quality matters because RAG picks specific URLs to retrieve, and index-friendly structure, including schema, speed, and open crawler access, can push a comparable page ahead of a slower, messier one.