Retrieval-Augmented Generation, or RAG, retrieves relevant information and supplies it to a model before it answers.
RAG combines retrieval with generation so a model can use information supplied at runtime.
A RAG system searches a source collection, places selected results into the model context, and asks the model to answer from that context. The source may be documents, a database, or another trusted system.
Think of it like asking a librarian before writing an answer.
The librarian finds relevant pages, and the writer uses those pages to answer. The analogy covers the retrieve-then-generate flow, not human-like understanding.
Input: a question about account settings
AI / System: retrieves relevant help articles
Output: an answer grounded in those articles
Input: a question about company policy
AI / System: searches approved internal documents
Output: a cited answer for the employee
RAG: supplies information at runtime.
Fine-tuning: changes learned model behavior through training.
RAG: retrieves information and uses a model to generate an answer.
Search: returns matching results without necessarily generating an answer.
RAG: gives the model selected evidence.
Guaranteed truth: cannot be provided by retrieval alone.
Explore next
RAG retrieves relevant evidence and gives it to the model before generation.