September 8, 2026 · 8 min read

What Is Retrieval Augmented Generation

A language model on its own is a closed book. It answers from what it saw in training, cannot see your documents, and invents a clean, confident answer when it reaches the edge of what it knows. Retrieval augmented generation, named by Patrick Lewis and his co authors in the 2020 paper Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks, adds one step: search. Index your documents, retrieve the passages that match a question, place them in the prompt, and let the model answer from them and cite the source. What RAG is, the four steps every system runs, and why it beats fine tuning when the facts move and the answer has to name its source.