Glossary
RAG (retrieval-augmented generation)
Retrieval-augmented generation (RAG) is a method where an AI system first searches a trusted source, such as company documents or a database, for relevant passages, then gives them to a language model so its answer is based on that material.
What Is RAG (Retrieval-Augmented Generation)?
In short A RAG pipeline has two halves.
Retrieval prepares your content in advance: it splits documents into chunks, turns each chunk into an embedding (a numeric representation of meaning) and stores them in an index, often a vector database. At question time it finds the chunks closest to the question. Generation then passes those chunks to the language model with the question and an instruction to answer from them, often with citations.
RAG fixes two limits of LLMs without retraining the model: the training cutoff and the lack of private knowledge. Update a document and the next answer reflects it, and answers can point to their sources. The term comes from a 2020 research paper, Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, presented at NeurIPS 2020.
RAG Example
For example, a company loads 300 pages of product docs, 40 case studies and its pricing FAQ into a RAG system. When a prospect asks whether the product integrates with NetSuite, the system retrieves the two relevant doc pages, and the rep's assistant drafts a reply that quotes them with links in about 10 seconds.
RAG in Practice
Sales teams meet RAG in tools that answer from their own material: a rep asks how the product handles single sign-on and gets the answer from the security docs, or an email assistant pulls the case study that matches a prospect's industry. Answer quality depends mostly on retrieval. Stale, duplicated or badly split documents produce wrong answers no matter how good the model is.
FAQ
What does RAG stand for in AI?
Retrieval-augmented generation: the AI retrieves relevant material from a knowledge source first, then generates its answer from that material.
What is the difference between RAG and fine-tuning?
RAG adds knowledge at question time by retrieving documents, with no retraining. Fine-tuning changes the model's parameters by training it on new examples. RAG suits facts that change; fine-tuning suits style and format.
Does RAG stop hallucinations?
It reduces them by grounding answers in retrieved sources that people can check, but it does not remove them. If retrieval returns the wrong passage, the model can still give a confident wrong answer.
What are the main parts of a RAG system?
A store of your content, usually an index or vector database; a retriever that finds the relevant passages; and a generator model that writes the answer from them.
The new Ken opens to the waiting list in batches.
Related
- GTM Glossary
Every GTM term, grouped by topic.
- LLM (large language model)
- Prompt engineering
- AI agent
- Model Context Protocol (MCP)
- Data enrichment
- Ken features
Sources
| Source | Checked |
|---|---|
| arXiv: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (2020) | |
| NeurIPS 2020 Proceedings: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks | |
| AWS: What is RAG? |
Changelog
- Page published.