Glossary

Grounding (Web Search)

Grounding is an AI assistant's use of live retrieved sources — usually web search — to answer, instead of relying on training memory alone. A grounded answer reflects what current pages say and can cite them; an ungrounded one reflects a model's frozen, dated impression.

Grounding is the single most important methodological detail in AI-visibility measurement. Consumer apps — ChatGPT, Gemini, Perplexity — search before answering local questions. An API call without search enabled measures a different, weaker system, and produces misleading results like assistants "refusing to name businesses" when the real product names six with citations.

We learned this on our own data: an early ungrounded run showed refusal rates up to 100%, which described the API configuration rather than any market. Everything published in our research uses grounded assistants exclusively, because that is what buyers actually use.

The failure was instructive because it was so plausible. An ungrounded model asked to name the best dentist in Austin has been trained to decline — it knows it cannot verify current local facts, and refusing is the safe behaviour. Read without knowing search was off, that looks like a finding about AI and local business: assistants won't recommend anyone. It is actually a finding about a checkbox. A great deal of published commentary on this subject rests on the same mistake.

Grounding is also what makes the channel winnable rather than a lottery. If assistants answered only from training memory, the businesses named would be whichever ones happened to be prominent when the model was trained, and nothing you did afterwards would move them until the next training run. Because retrieval happens at answer time, the evidence available today decides the answer, and improving that evidence changes what the assistant says within weeks.

For anyone evaluating an AI visibility report, this is the first question to ask and the easiest to check: was web search enabled, and can you see the sources? A grounded answer names where it looked. An answer with no citations either came from memory or from a tool that is not telling you which.

Related terms
  • Retrieval-Augmented Generation (RAG) Retrieval-augmented generation is the technique behind grounded AI answers: before responding, the system retrieves relevant documents — search results, web pages, database records — and generates its answer from them.
  • AI Citations AI citations are the sources an assistant links or names to justify its answer — the pages it actually read before recommending businesses.
  • Knowledge Cutoff A knowledge cutoff is the date after which an AI model's training data ends.

See where your business stands.

The free AI visibility report shows what ChatGPT, Gemini, and Perplexity say when buyers ask about your market — with the sources behind every answer.

Get the free report All terms

No charge and no call required — the report arrives by email.