Glossary
Grounded Citation
A brand mention an AI engine produces during a live web search — retrieved right now, not recalled from training data.
Definition
A grounded citation is a brand mention an AI engine produces while performing a live web search: the engine fetched a page during the answer and cited the brand from what it retrieved. The opposite — an ungrounded mention — comes from the model's training data, a snapshot that may be many months old.
The two look identical to the person reading the answer. They behave completely differently for the brand trying to improve: one responds to today's site, the other to history.
Grounded vs training-data recall
| Dimension | Grounded citation | Training-data mention |
|---|---|---|
| Source | Live retrieval during the answer | Model weights from training |
| Freshness | What the engine can fetch now | Snapshot, possibly months old |
| Responds to fixes | Within days of a crawl | Only after a retrain you cannot schedule |
| What it verifies | Your site is retrievable and citable today | Your brand was in the training corpus |
This is why grounded monitoring is the feedback loop for GEO work: if a schema or content fix lands, grounded results are where the change can first become visible — and where AI crawler analytics confirm the engines actually re-read the page.
Frequently asked questions
- What is a grounded citation?
- A grounded citation is a brand mention an AI engine produces while performing a live web search — the engine retrieved a page right now and cited the brand from it. It contrasts with training-data recall, where the engine repeats what it memorized months earlier.
- Why does grounded vs ungrounded matter?
- Because they respond to your work on different clocks. Grounded results reflect what the engine can retrieve today, so content and schema fixes can influence them within days. Training-data mentions only change when models are retrained, which you can neither schedule nor observe directly.
- How does ApexGEO monitor grounded citations?
- ApexGEO runs grounded monitoring on ChatGPT, Gemini, Claude, and Grok — engines whose APIs expose a live-search mode — and records grounded results separately from ungrounded ones, so a measurement never mixes what an engine retrieves now with what it memorized during training.
- Is a grounded citation better than a training-data mention?
- They answer different questions. A training-data mention shows the model absorbed your brand historically. A grounded citation shows your site is retrievable and citable right now — which is the signal you can actually act on, and the one that verifies whether a fix worked.
See your grounded citations
ApexGEO runs grounded monitoring on the engines that expose live search, and reports grounded and ungrounded results separately — see the methodology for how runs are scheduled and recorded.
7-day free trial, no card required — on the self-serve plans.