What is RAG (retrieval-augmented generation)?

What is RAG?

RAG stands for retrieval-augmented generation. A regular AI answers from what it learned in training, which means it can be out of date, generic, or just wrong about your business. With RAG, the AI takes the question, searches a stack of documents you gave it (a folder of policies, a Notion workspace, last quarter's contracts), pulls the few relevant snippets, and writes its reply using those snippets as the source.

A common confusion: RAG is not the same as "training the AI on your data." The AI does not memorize your docs. It looks them up fresh every time. Add a new policy PDF on Tuesday and the agent answers from it on Wednesday, with no retraining.

When you'd encounter it

A support agent that answers customer questions from your help center and points to the article it used. A sales agent that quotes prices from a Google Drive playbook. An internal HR bot that reads from the current benefits doc, not last year's version. The value: the answer is grounded in your stuff, and the agent can show you which doc it came from.

For engineers

Under the hood, RAG embeds your docs into a vector store, runs top-k similarity search against the user query, and injects the retrieved chunks into the prompt. Chunking strategy and reranking are where most production RAG systems live or die.