RAG (retrieval-augmented generation)
Japanese RAG（資料を探して付け足し、答えさせる方法）
RAG is a way of having an AI first look up related material, such as company documents or web pages, and attach it to the question before answering. It lets the AI answer from current or your own information without retraining the model.
RAG is like letting an AI take an open-book exam. When a question comes in, the system first searches material such as company documents or web pages for the relevant parts, then adds that text to the question before passing it to the AI. The AI builds its answer not only from what it learned in training but from the material placed in front of it. You can get answers based on new information just by changing the material, without retraining the model, and the AI can show which documents it used.
Watch out RAG does not make every answer correct. If the material it finds is outdated or off topic, the answer can be wrong too, so when a source is shown, open the original and check it. When you connect internal documents, set it up so they are used only for questions from people allowed to see them. When a file you uploaded changes, upload the new version to keep it current.
Source: Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks · What is RAG? - Retrieval-Augmented Generation AI Explained · RAG and Generative AI - Azure AI Search (checked 2026-10-26)