Grounding
Grounding ties an AI's answer to specific sources you supply — your files, pasted text, or web results — instead of its fuzzy training memory.
Definition
Grounding means handing the model the actual source material — your documents, text you pasted, or live web results — and instructing it to answer only from that material. The answer is anchored to evidence you can see and check, not to the model's general recollection. It is like the difference between a student answering from memory and one allowed to quote directly from the open textbook in front of them.
Why it matters
An ungrounded model answers from a blurry memory of everything it once read, which is exactly where confident-but-wrong replies come from. Grounding keeps it on the facts you provided and pairs naturally with citations, so each claim can be traced back to its source. In AskClear's Facts mode this is the default: your answer about a contract quotes the contract, and you can see the exact line it came from.