Grounding

Grounding is a design approach that ties an AI's answers back to real documents or data, making it "cite its sources." It's like writing on an exam that the answer comes from "page X, line Y of the textbook."

AI sometimes makes up statements that sound plausible but are wrong — this is called hallucination. Grounding is a key strategy for reducing it. Instead of letting the AI generate text freely, restricting it to answer only based on given documents leaves less room for it to invent things.

In practice, RAG (Retrieval-Augmented Generation) is the most common way to implement grounding — when a question comes in, relevant documents are retrieved first, and the AI answers while reading those documents. When an article about enterprise AI adoption mentions "applying grounding," it means the system was restricted to answer based on the company's own documents to boost reliability.

"Grounded in internal company documents to reduce hallucination" — this means the AI was designed to answer based on internal materials, improving accuracy.