Glossary

Grounding

Whether an AI answer draws on live web retrieval (a grounded, or retrieval-augmented, answer) versus fixed training-time knowledge. Changes what optimization work can influence, and how fast.

Definition

Grounding refers to whether an AI system's answer draws on live web retrieval, current, indexed content pulled in real time, rather than relying only on knowledge fixed at training time. A retrieval-augmented (RAG) answer can be influenced by a page published this week; an answer drawing purely on training data will not reflect that page until a future training cycle.

Some engines ground most answers by default (Perplexity treats nearly every query as a live web search); others blend grounded and ungrounded answers depending on the query. Where it is knowable, scores on this site are segmented by whether the underlying answer used live web grounding, since the two behave differently enough to matter.

Same ruler

See how this is measured

This term is part of the same declared methodology every score on this site follows: which prompts, which engines, how many runs.

Read it in the methodology
How Does ChatGPT Choose Which Sources to Cite?
All glossary terms

Want to see where your own brand stands?

One-time report, no subscription, delivered in days.

Get the full report