What is retrieval-augmented generation (RAG)?
Retrieval-augmented generation (RAG) is a method in which an AI system first searches a set of documents or the web for relevant passages, then writes its answer using what it found. Retrieval is the searching step. AI search tools work this way.
Last updated
In more detail
A language model on its own answers from memory of its training text, which may be old or incomplete. Retrieval gives it something current to read. When a question arrives, the system fetches the most relevant passages from an index, such as a company's help articles or pages from the live web, places them in front of the model, and the model composes a reply from them, often with links to the sources. AI search products and many customer-service assistants work this way. For marketers, it explains why some pages are quoted and others are not: only content that can be retrieved can be used. Pages need to be crawlable, clearly organised and written in self-contained passages that still make sense when lifted out. The common mistake is assuming retrieval removes all errors. If the retrieved page is wrong or out of date, the answer will be too, and the model can still misread a good source.
Related terms
- GroundingGrounding is the practice of tying an AI model's answer to specific, checkable sources, such as retrieved web pages or supplied documents, so that its statements are based on evidence instead of on the model's memory alone.→
- AI citationAn AI citation is a reference an AI assistant or AI search feature attaches to its answer, naming or linking the web page that a statement came from.→
- Large language model (LLM)A large language model (LLM) is an AI system trained on very large amounts of text to predict and generate language.→
- AI crawlerAn AI crawler is a piece of automated software run by an AI company that visits websites and reads their pages, either to gather training material for models or to fetch content for answers in AI search and chat.→
More in AI search
Do it with Wonib
Put the words into practice
Wonib learns your brand once, then makes the posts, pictures, copy, decks and documents in it, checked against the brand before you see them. Start free with 1,000 credits, no card needed.
Last updated