What is retrieval-augmented generation (RAG)?

By the Variask team · Updated

Definition

Retrieval-augmented generation (RAG) is a way of building AI answers where the system first retrieves relevant documents or web pages, then has the language model write the answer using them.

Also called RAG

How RAG works

  1. 1Retrieve: find the passages most related to the question, using search, vector search or both.
  2. 2Augment: add those passages to the prompt the model sees.
  3. 3Generate: the model writes an answer based on them, often with citations.

AI search engines such as Perplexity, ChatGPT search, Copilot and Google AI Overviews all follow this pattern in some form.

Why it matters for visibility

In a RAG system, the answer can only use what was retrieved. If your page is not among the retrieved passages, you are not in the answer, however good your product is. Being retrievable, readable and quotable is the core of AI search optimization.

What it means for brands

  • Be findable in the searches the engine runs, not only in the question the user typed.
  • Make each passage clear and self contained so it survives being lifted out of the page.

Questions

Retrieval-augmented generation (RAG):
common questions

No. Fine tuning changes the model with extra training. RAG leaves the model as it is and gives it fresh material at answer time.

No. Many answers come only from training data. RAG is used when the engine searches or has access to a document set.

Related terms

Keep exploring

How AI search works

Grounding

Grounding is when an AI model bases its answer on specific sources, such as live search results or a set of documents, instead of relying only on what it learned in training.

Read

How AI search works

Vector search

Vector search finds content by meaning instead of exact words. Text is turned into lists of numbers called embeddings, and the search returns the items whose numbers sit closest to the question's.

Read

How AI search works

Query fan-out

Query fan-out is when an AI engine splits one question into several related searches, runs them together and blends the results into one answer.

Read

How AI search works

Content chunking

Content chunking is the splitting of a page into smaller passages, so an AI system can store, find and quote the most relevant part instead of the whole page.

Read

How AI search works

Agentic search

Agentic search is when an AI assistant plans and runs its own research before it answers: it breaks a question into steps, searches several times, reads pages, compares what it finds and only then writes a reply.

Read
Full glossary