Retrieval-augmented generation (RAG)

Giving a language model relevant documents at question time, so it answers from them instead of only from its training data.

Last updated September 28, 2026.

Definition

What Retrieval-augmented generation (RAG) means

Retrieval-augmented generation retrieves documents related to a question, from a search engine, a database, or an index of your own text, and puts them in the model's prompt. The model then answers from those documents and can cite them.

RAG addresses two limits of language models: their knowledge stops at a training date, and they cannot point to where a claim came from. The answer is only as good as what is retrieved, so the retrieval step, and knowing the date and source of each document, matters as much as the model.

Clair's Google Search API can serve as the retrieval step for current web results, and the news article endpoints return full text with URL and publish time for an index.

In the data

How it appears in a response

GET /v1/article?engine=ap_news&url=https%3A%2F%2Fapnews.com%2Farticle%2Fhurricane-polo-mexico-nolo-hawaii-a036e08f59c3f3842aaba2ee69bd9d37 · article.paragraphs

[
  "Hurricane Polo dumped heavy rain on Mexico’s Pacific coast."
]

Article text as paragraphs, ready to chunk for an index. From a recorded Associated Press API response.

Subscribe to the API you need.

Each Clair API is its own monthly request quota. Unused requests expire. Every API has a free tier to try it.

Get API Key →Compare plans

Glossary