What Is RAG?
Retrieval-Augmented Generation (RAG) is a technique that lets an AI model pull in relevant information from an external source — a document set, a database, a knowledge base — before generating its response, rather than relying purely on what it learned during training. In practice, this means the model can answer accurately about your own specific documents or recent information it was never trained on.
Why It Matters for Accuracy
A model's training data has a cutoff date and doesn't include your organization's private documents. RAG addresses both gaps: it retrieves relevant, current, or private content at the moment of a query and gives the model that content as context, meaning), the response can be grounded in real, verifiable source material instead of the model's general training.
How It Works, Conceptually
A RAG system typically works in three steps: your documents are broken into chunks and converted into numerical representations (embeddings) stored in a vector database; when you ask a question, the system finds the most relevant chunks based on similarity to your query; those chunks are then given to the model as context alongside your question, so its answer is grounded in that specific material.
Where You'll See It
RAG underlies most "chat with your documents" features and enterprise AI assistants that need to answer accurately about internal knowledge. Tools like NotebookLM (in our Platforms directory) are built around this pattern specifically.
Related Pages
Frequently Asked
Is RAG the same as fine-tuning a model?
No — fine-tuning changes the model's underlying weights through additional training; RAG leaves the model unchanged and instead feeds it relevant external content at the moment of a query.
Why not just put all my documents in the prompt directly?
For small document sets this can work, but most real document collections are far larger than any model's context window allows, which is exactly the problem RAG's retrieval step solves.
Does RAG guarantee an accurate answer?
No — it significantly improves grounding, but the model can still misinterpret retrieved content or, if the retrieval step returns weak or irrelevant results, produce an inaccurate answer regardless.
What's a vector database's role in RAG?
It stores the numerical representations of your document chunks and is what makes the fast similarity search step in retrieval possible; see our Vector Databases page for more detail.