Glossary
What is retrieval-augmented generation (RAG)?
Retrieval-augmented generation (RAG) is a way of making a language model answer from specific documents: the system first retrieves relevant passages from a search index or database, then gives them to the model to write the answer. AI search engines use it to answer from the web with citations.
How it works
- The question is turned into one or more search queries.
- A retriever finds relevant documents or passages.
- The best passages are placed in the model's context window.
- The model writes an answer based on them, often citing which passage supported each claim.
Why it matters for websites
Perplexity, ChatGPT search, Google's AI features and many assistants work this way over the web. Two things decide whether your page contributes: whether it's retrieved (search ranking, crawlability, relevance) and whether a passage from it is clear and specific enough to use. Pages with direct answers, numbers and headings are easier to use.
RAG and accuracy
RAG reduces made-up answers because the model has sources in front of it, but it can still misread or combine sources wrongly. Consistent facts across sources make errors less likely.
Example
A user asks an assistant how many startup directories give dofollow links. The retriever pulls a statistics page and two blog posts; the model answers with the statistics page's number and cites it, because it stated the figure with a clear method.
FAQ
Is RAG the same as fine-tuning?
No. Fine-tuning changes the model's weights; RAG gives it documents at answer time.
Do all AI assistants use RAG?
Most do when search or document tools are on. Without them, answers come from training data.
How do I get my pages retrieved?
Rank in search for the questions, allow the relevant crawlers, and write clear, specific sections.
Sources
- Lewis et al.: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (arXiv)
- Anthropic: Contextual retrieval