KnowledgeMeasurement
How do you measure AI visibility reliably?
AI visibility is measured by asking the same buyer questions daily, without a login and across several AI engines, and evaluating the share of answers in which a company is recommended: the recommendation rate. A single query says almost nothing; only the trend over weeks is robust.
Why a single query says nothing
AI systems do not keep a ranking on which your company holds a fixed position. Every answer is generated anew, and the same question returns a different selection of typically three to five providers on every repetition. A report showing a single measurement per question shows you a snapshot of a system that will answer differently on the next query.
Visibility is still measurable, but only as a measurement series: the same questions, the same conditions, repeated over weeks and compared against the same competitors.
The measurement protocol
This is how we measure in every GEO retainer at Carigiet GEO:
50 buyer questions per service area
The questions reflect what real buyers ask, not what would be asked internally. They are developed from the company’s knowledge base during onboarding and reviewed by the client. How such a set is built is shown in developing buyer prompts.
Daily across 6 engines
Every question runs every day on ChatGPT, Claude, Perplexity, Gemini, Copilot, and Google AI Overviews: 300 runs per day. Measurement runs on Peec AI; the Claude model used is Claude Haiku.
Without a login and without memory
Queries run stateless so the measurement does not influence itself. Otherwise you end up measuring your own history.
Weekly comparison against a defined competitor set
Evaluation happens weekly, always against the same named competitors. Only then is a change a change and not noise.
Mention, citation, recommendation, and the recommendation rate
Four terms that often get mixed up in reports:
Mention
Your brand appears in the text of an AI answer.
Citation
The answer references one of your pages as a source.
Recommendation
The system proposes your company as an option for the buyer question asked.
Recommendation rate
The share of repeated identical queries in which your company is recommended. That is the metric we measure progress by, not the order within a single answer, which is not stable across repetitions.
What is a good value?
For a Swiss SME in a competitive B2B field, 30 to 40 percent is a strong result. That is a reference value, not a promise: it follows from the mechanics that only three to five providers are named per answer, with usually dozens of competitors contending for those spots. A company recommended in every third repetition of the same buyer question is stably part of the shortlist.
The entry point into the measurement series is a baseline. The free AI visibility analysis sets it with a first structured snapshot; the picture becomes reliable once the same questions keep running over the following weeks.
Measurement shows the state, not the cause. Which levers actually move the recommendation rate in ChatGPT is set out in what ChatGPT optimization covers.
