Measuring Visibility When There Is No Ranking
AI answers have no position 3. How to survey visibility anyway – with a sample you can repeat, and without claiming precision the data cannot support.
Boaz Lichtenstein
Google Search Console’s generative AI performance report counts appearances of links to your site in AI Overviews and AI Mode. You can separate text-based and multimodal searches. The report does not show which question someone asked or which AI link they clicked. A brand can also appear in an answer without a link to its website. A repeatable sample lets you investigate such mentions. How to read the AI report
The survey you can repeat
The only value of such a measurement lies in comparison with yourself. For that, four things must be fixed and must not change:
- The question list. Ten to twenty questions phrased the way a customer would ask – not the way an SEO would. Exactly this list, every time.
- The models. Two or three, noted with their version. Swapping a model between two measurements makes the series worthless.
- The cadence. Monthly is enough. More often mostly measures noise.
- What gets counted. Brand mentioned, domain linked, position within the source list. Three numbers, not a feeling.
The LLM Brand Check helps you hold the input and model constant across runs. Even with identical settings the answers can vary.
Variance is the normal case
The same question asked three times yields three answers. That is not a defect, it is the design. Two consequences:
Repeat instead of asking once. Three runs per question, then note the hit rate – “mentioned in 2 of 3 runs” is a statement, “mentioned” is not.
Report a range, not a decimal. Turning 17 queries into “34.7 percent visibility” invents a precision the data does not support. “In roughly a third of cases, with clear variance between models” is more honest and just as usable for decisions.
What this measurement does not say
It does not say whether it produces revenue. It does not say how many people saw the answer. It does not say whether your text was the reason for the mention. Turning it into a channel report sells a guess as a number.
What it gives you is an early-warning value: if mentions fall across three measurements, something changed – in you, in the models, or in the competition. That is a reason to look closer, not a result.
The effort, honestly calculated
Twenty questions times three runs times three models is 180 queries a month. That is an hour of work or a small script – and it is more substance than any external study on “AI visibility”, because it is imprecise about your brand rather than about the market.