Definition
Grounding rate measures how often an AI engine answers from retrieved web sources instead of from parametric knowledge—the facts baked into the model during training. It is a property of the prompt and the engine, not of your brand, but it determines which optimization levers can work at all.
The distinction is decisive for GEO. When an answer is grounded, retrieval is running, so being crawlable, structured, and retrievable can put your content in the response this week. When an answer is ungrounded, the model is drawing on training data, and no amount of on-site optimization changes the output in the short term. Influencing ungrounded answers is a slower game of broad, consistent web presence—the substance behind brand depth and share of model.
Grounding rate varies sharply by prompt type and engine. Time-sensitive, local, comparative, and product questions trigger grounding queries far more often than definitional or conceptual ones. Retrieval-first products like Perplexity ground nearly everything, while general assistants ground selectively based on their own triggering logic. Measuring the rate across your prompt set tells you what proportion of your target queries are winnable through content work.
A related quality dimension is grounding depth: whether an answer's individual claims are genuinely attributable to the cited sources, or whether sources are appended to text the model largely generated on its own. Weakly grounded answers cite less stably over time, so prompts with high grounding rates and strong source-claim alignment are the most durable places to compete. Tracking citation rate alongside grounding rate gives a fuller picture.
Examples of Grounding Rate
- A team finds 70% of its monitored prompts trigger live retrieval, so content work can influence most of its target queries.
- A definitional prompt is answered entirely from model memory, so the brand's new landing page has no effect on the response.
- Comparison and pricing prompts show much higher grounding rates than conceptual ones, shifting content priorities toward buying-stage queries.
- An analyst separates grounded from ungrounded prompts before reporting citation rate, since ungrounded prompts can never produce a citation.
