What is grounding?
Grounding is connecting an AI answer to source material retrieved for that question, so statements can be checked against pages rather than left to the model alone. Citations are the visible form of grounding when an engine shows them. Grounding lowers the chance of invention. It does not make every sentence true.
Why it matters for AI visibility
Ungrounded answers can sound confident and still invent prices, features, or customers. For a brand, that is an AI visibility problem and an accuracy problem: you may be described wrongly, or a rival’s page may be the only evidence on screen. Grounding matters because it is the mechanism that lets an engine attach a claim to a URL you can audit. You influence it indirectly. Publish pages that state checkable facts, keep them crawlable, and earn mentions on sources the engine already retrieves. You do not get a control that forces a model to ground on your domain. When you review an answer, separate three outcomes: a grounded citation to you, a grounded citation to someone else, and an unlinked claim with no source. Each one implies a different fix.
Example
An assistant says your plan includes a feature you do not sell, with no link. That claim is ungrounded. A later answer links to your pricing page and repeats the prices printed there. That claim is grounded in your page, which you can correct if the page is wrong or outdated.
Related terms
Retrieval-augmented generation is the pattern of fetching documents and then writing from them. Hallucination is the failure mode grounding is meant to reduce. An AI citation is the link a person can click when grounding is shown.