What's the difference between grounding and RAG?

Grounding is the goal, tying an answer to a specific source a person could go check, and retrieval-augmented generation, RAG, is one way to reach it: retrieve relevant text and put it in the model’s context before it answers. They get used interchangeably because RAG is the most common grounding technique, but it isn’t the only one. A tool call to a live database, a citation against a document the user pasted directly, or a structured lookup all ground an answer without retrieving or ranking anything.

The relationship only runs one way. Every RAG answer is an attempt at grounding, but retrieval doesn’t guarantee the attempt landed: context presence, groundedness, and faithfulness are the three separate ways it can still fail, from retrieval never finding the source to the model ignoring what it found once it had it. Grounding is the property you actually want; RAG is one common architecture for getting it, not a synonym for having it.