Definition
Retrieval-augmented generation, or RAG, combines a search step with a language model. When a question arrives, the system retrieves relevant documents or passages, often by searching a web index or a vector database, and gives them to the model as context. The model then writes its answer using that material and may cite it.
For example, when a user asks an assistant whether Closedesk integrates with popular accounting tools, the system might retrieve Closedesk's integrations page and a recent help article, then answer from those pages. Without retrieval, the model would rely only on training data, which may be outdated.
Why it matters
RAG is why live pages can influence AI answers without waiting for a model to be retrained, and why citations exist at all. For brands, it means content that is crawlable, indexed, current and easy to retrieve has a direct route into answers. It also means retrieval quality matters: if the system cannot find or match your page to the question, even excellent content will not be used.
What to do
- Keep key facts on crawlable, indexable pages rather than in PDFs or behind logins.
- Write sections that answer one question each, so passages can be retrieved on their own.
- Check which of your pages assistants cite to understand what retrieval systems find.
Common questions
What is the difference between RAG and grounding?
RAG is a system design: retrieve documents, then generate an answer from them. Grounding is the broader goal of tying an answer to verifiable sources. RAG is one common way to achieve grounding. In practice, the terms are often used together when discussing cited AI answers.
Do all AI assistants use RAG?
Many assistants use some form of retrieval, especially for questions that need current information, but not for every answer. Whether retrieval runs can depend on the question, the product and user settings. When it does not, the answer comes from the model's training data alone.