Grounding
Grounding means tying the answer of an AI model to supplied, verifiable sources: the model is only supposed to state what can be supported by the documents provided.
In practice
The most common grounding technique is RAG: before answering, matching passages are retrieved from a knowledge base and handed over together with the question, often including a source reference. This makes answers traceable – anyone in doubt can look up the cited passage.
It is not a guarantee. Even with sources in front of them, models invent content. When merely summarising supplied texts, even the best models produce around two to three percent of fabricated statements (Vectara leaderboard, data as of May 2026). A Stanford study in 2024 found 17 to 33 percent faulty answers in RAG-based legal research services, depending on the provider. Human review therefore remains part of the process.
It is not a guarantee. Even with sources in front of them, models invent content. When merely summarising supplied texts, even the best models produce around two to three percent of fabricated statements (Vectara leaderboard, data as of May 2026). A Stanford study in 2024 found 17 to 33 percent faulty answers in RAG-based legal research services, depending on the provider. Human review therefore remains part of the process.