grounding
tying an AI's answer to real records instead of its memory
An AI model can write fluent answers that are wrong. Grounding means the system first fetches real data, such as a product list or an order record, and the answer is built from it. Strong grounding checks in code that every figure shown came from those records, so a made-up price never reaches the customer.
Think of it as
A shop assistant who checks the price tag before quoting a price, instead of saying what they remember.
Ideas that use this term