What Is RAG in AI? Plain-English Definition

Description

Glossary What is retrieval-augmented generation (RAG)?

Retrieval-augmented generation (RAG) is a way of building an AI assistant in which the system first retrieves the relevant passages from a chosen set of documents, then gives those passages to a language model to write the answer from. The model answers from your material, not only from what it learned in training.

Talk to a technology expert Last updated 2 October 2026

How it works

Before any question At each question

Your documents Question

| |

split into passages 1. search the index

stored in a 2. best-matching passages

searchable index |

3. question + passages -> language model

4. answer, with the sources it used

The term comes from a 2020 research paper by Patrick Lewis and co-authors, which combined a language model with a searchable store of documents so that the model could draw on knowledge outside its own training.

A two-line example

Question: "Can I return an item bought in the sale?"

Answer: "Yes, within 7 days if it is unused. Source: Returns Policy, section 3."

The model did not know the shop's returns policy. The system found section 3, passed it to the model with the question, and the model wrote the reply from it. The policy text here is invented for the illustration.

Why it matters when you commission software

RAG is the usual way to make an assistant answer about your business, because it needs no retraining and it can show where an answer came from. The quality of the result depends far more on the documents and the search step than on the model. Ask:

Question to ask

Why

Which documents will it use, and who keeps them current?

It repeats what the documents say, including what is out of date.

Does each answer show its source?

Without one, nobody can check it.

What does it say when nothing relevant is found?

"I don't know" is the correct behaviour, and it has to be built.

Can it show a user documents that user is not allowed to see?

Access rules must apply to the search step too.

How is it tested?

A fixed set of real questions with known answers, re-run after every change.

When you do not need it: if the answers fit on one page of text, that page can be given to the model directly and no retrieval is needed. WavX builds these systems as RAG development ; if you already have an assistant that answers wrongly, start with AI chatbot giving wrong answers.

Related terms

Large language model (LLM): the AI model that writes the answer.

Embedding: a numerical representation of a passage that lets the system search by meaning, not only by matching words.

Vector database: the index where embeddings are stored and searched.

Fine-tuning: changing the model itself with further training, a different technique; see RAG vs fine-tuning vs prompt engineering.

Frequently asked questions

Does RAG stop an AI from making things up?

It reduces it; it does not remove it. The model can still misread a passage, and if the search step finds the wrong passage the answer will be wrong with a source attached. A sound build tests for both and tells the user when nothing relevant was found.

Is RAG the same as training the AI on my data?

No. The model itself is not changed. Your documents are stored in a searchable index, and the relevant parts are handed to the model at the moment of each question. Updating the assistant means updating the documents.

What kind of content works with RAG?

Text the answer can be found in: policies, manuals, product information, past support replies, contracts. It works poorly when the knowledge is not written down anywhere, or when the documents contradict each other.

Sources

Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (arXiv 2005.11401, NeurIPS 2020) · read 2 October 2026

AWS: What is Retrieval-Augmented Generation? · read 2 October 2026

Related

RAG development

RAG chatbot development cost

AI development cost calculator

Build your own software — your way, your pricing.

WavX Solutions is here to create your own software in a fully custom way, built exactly how you work — with a pricing model that fits your business. Connect now and let's build it.

Contact Now helpwavx@gmail.com