RAG: give an answer the right evidence
Explain how retrieval helps answer questions about changing documents.
In plain language
Retrieval-augmented generation first finds relevant material, then supplies that material to a model as context for an answer. Search may use keywords, vectors or a combination. A vector database is not a requirement of every RAG system.
The quality of the answer depends on what was retrieved and how the model uses it. Stale, irrelevant or unauthorized content can cause problems. Evaluate retrieval separately from the final answer, and keep citations linked to the supporting passages.
A support question asks about a recently updated feature. Search the current product manuals, select the relevant passage, and have the assistant answer with its source. Manually pasting a passage shows added context; automated selection is the retrieval step.
Try it yourself
For a question about a changed policy, name the retrieval step and how you would verify the answer against the source.
Go deeper · for developers and architects
Changing chunk size, overlap or ranking can help particular workloads. Diagnose failures using test queries before choosing a change. More retrieved text is not automatically better.
Check your understanding
3 related questions, with an explanation for each answer.
Explore the original sources
Microsoft RAG overview ↗Report an issue
Tell the editors what needs correcting. Reports are reviewed and do not automatically remove content. Do not include personal or confidential information.
Finished this lesson?
Completion records your choice, not a measured skill level.