Gemini has no crawler you can allow by name
This is the first thing a "Gemini visibility tracker" has to get right. Google documents Google-Extended not as a crawler but as a usage control: it decides whether content Google has already crawled may be used to train Gemini models and to ground answers in Gemini Apps and Vertex AI. The fetch itself is done by Googlebot. Google states explicitly that this control does not affect inclusion or ranking in Google Search. Two consequences follow. Blocking Google-Extended keeps you in search while removing you from grounded Gemini answers — a bounded, specific cost, not the vague sitewide one people assume. And because it governs usage rather than fetching, nothing appears in your server logs when it is honoured; there is nothing on your side to verify. The documentation is at Google's crawler reference.
A second documented agent, Google-CloudVertexBot, crawls sites only at the site owner's own request when building Vertex AI Agents. It is irrelevant unless you are building an agent on your own content.
So tracking happens on the answer side, not the log side
Since you cannot watch Gemini fetch anything, the only place to measure is the answers themselves. That means the same discipline as for any answer engine: a fixed set of buyer questions with no brand names in them, each run more than once because generated answers vary between identical runs, and three outcomes recorded separately — mentioned (the brand is named), cited (a URL of yours is attached as a source), recommended (the answer suggests you for the question). Report them as a rate over the set. There is no position inside an answer, so anything presented as a Gemini "rank" is the order brands happened to appear in one sample.
What a Gemini tracker should record for every answer
| Field | Why it matters for Gemini specifically |
|---|---|
| Question, engine, date, run number | Without these you cannot tell a detection bug from a real change — and both will happen |
| Whether the answer was grounded at all | A Gemini answer with no sources attached is a different event from one that cited a competitor; do not pool them |
| Cited URLs, taken from the answer body only | Sources retrieved but not attached are not citations |
| Your Google-Extended policy at the time | If you change the directive, the series restarts; record it next to the run so a later reader can explain a step in the chart |
| Failed runs, left blank | A run that did not complete is not 0% — rendered as zero it turns a flat trend into a fake decline |
Reading a result
- Named but never cited: the entity is known; your pages are not being used as sources. Look at whether the page states facts in a form that can be lifted out.
- Cited but never recommended: your page is reference material while the answer sends the buyer elsewhere. That usually points at corroboration outside your own domain.
- Absent while competitors are named: check whether the page is actually indexed in Google Search first — grounded answers draw on the same index, and a page Search has not indexed cannot be grounded on.
The measurement rules are spelled out in brand share in AI answers; the general method in how to track brand mentions in ChatGPT applies to Gemini unchanged. For a first sample before building any of this, the free checks at tools.insightwonder.com are a starting point.