Glossary · AI Search

Grounding.

Also called: Retrieval-Augmented Generation, RAG

An AI model has two ways to answer: from memory (what it absorbed during training, which is frozen and can be out of date) or by grounding, going and fetching current sources at the moment you ask, then answering from those. Most modern AI search, ChatGPT with search, Perplexity, Google AI Overviews, is grounded: it retrieves live pages and cites them.

This is the whole reason a small business can show up in AI answers at all. A frozen model would only "know" big, famous entities from its training. A grounded system can pull your fresh, well-structured page into an answer about "electricians near Wauwatosa" today, even if you were nowhere in its training data. Grounding is the door a newcomer walks through.

The practical implication: being retrievable is step one. Your pages have to be crawlable by AI bots, current, and clearly relevant to the question, so that when the system reaches out to ground its answer, yours is among the sources it grabs and credits. Blocked or stale pages never get retrieved, so they never get grounded, so they never get cited.

A plain example

When Perplexity answers with live links it clearly just fetched, that is grounding. It did not "remember" those businesses; it retrieved them a second ago, which is why a brand-new, well-built page can appear.

Why this glossary exists. We define every term plainly, because an owner who understands the work makes better decisions, and asks sharper questions on the call. Ask one directly.
Call nowSee pricing