Grounding
What is grounding?
Grounding means tying an AI system's answers to specific, trusted information, such as your knowledge base, customer record or policy documents, so that what it says can be traced back to a source.
A grounded answer is one the AI could point to evidence for. In practice, grounding is achieved by giving the model the relevant facts at the moment it answers (for example through retrieval-augmented generation or by passing in the customer’s account details) and instructing it to use only that information.
Grounding matters in customer service because answers have consequences: refunds, eligibility, safety advice, account changes. It also makes AI auditable, since a supervisor or quality analyst can see which article or record an answer came from. Good grounding depends on good data: a single, current source of truth for knowledge and customer context, rather than several conflicting copies across systems.
Related: Retrieval-augmented generation (RAG), Customer context, AI hallucination.