Back to concepts
Retrieval-augmented generation (RAG)
What this means: An approach in which a system first retrieves relevant material and gives it to a language model as a basis for its answer. The answer then draws on retrieved material, but the model can still misinterpret it and the search can retrieve the wrong material.
ArchitectureStable concept
Sources and review
Reviewed September 10, 2026
Also known as
RAG, retrieval-augmented generation, retrieval-assisted generation
Explained in guides
- What are RAG and vector databases? Explained with a kitchen analogyHow can AI answer questions about your documents? Learn about RAG and vector databases through a simple kitchen analogy.
- How does AI search for information? Regular search vs. AI searchRegular search lists pages to choose from; AI search can combine sources into a direct answer. Learn how it works and why you should still check the source.
- Check the sources behind an AI answerLearn to check where a web-searching AI got each fact, whether the publisher and date are reliable, and what to do when sources disagree.
- Can you train AI yourself? — the basics of fine-tuningUnderstand fine-tuning in plain language: start with a use case and sample examples, not production data or assumptions.