Grounding

Grounding is the practice of anchoring an AI tool's answer in real, retrieved documents rather than in what the model remembers from training. When an answer is grounded, the tool has fetched specific pages — web results, a product feed, a database record — and composed its response from their contents, typically citing them. When it isn't, the tool is reciting from memory, which is where invented prices, discontinued products and confident errors come from.

Every major AI tool now grounds shopping answers because the alternative is unreliable, and that's the single best piece of news for an independent store: the tool reads your page as it is today. A grounded answer can only cite what it can read, though. Facts that live in images, in a JavaScript-rendered tab, or in a PDF are effectively absent; facts that are stated in the page text and confirmed in structured data are usable. Grounding is the mechanism that lets a page fix made this week appear in next week's AI answers — and the reason the "how to improve" advice throughout this glossary keeps returning to the same instruction: put the fact on the page, as text, and label it.