How Hypervisible measures AI visibility
The questions asked, the five AI engines, the Monday reading and how each figure is counted.
Every Monday, Hypervisible asks five AI engines the same set of questions, reads every answer in full, stores it with its sources and counts who is cited. Here is how each figure is obtained.
1. The questions
From a brand's industry, Hypervisible writes the questions a customer asks an AI before choosing: which is best, comparisons, prices, where to buy. These questions never name the brand or a competitor: they measure what a customer sees when searching without the brand in mind. A paid plan tracks 20 of them (5 in the free analysis), plus 5 questions that name the brand (reviews, comparisons, alternatives) to track its reputation.
Each question carries its demand: the monthly volume of Google searches and AI prompts on its topic. Demand ranks the questions, from most to least searched; it never changes the figures. Customers can add or remove questions; the set is then flagged as changed.
2. The five AI engines
ChatGPT, Gemini and Perplexity are queried with web search on. Google's two generated answers, AI Overviews and AI Mode, are read on live results. Questions are asked in the market's language (United States, United Kingdom, France, Germany). Answers come from Google and AI data; each one is stored in full, with its sources.
3. The Monday reading
AI answers vary from hour to hour. A daily reading would mistake that noise for real movement. Every Monday the same set is re-run on the same five AI engines, and the change is expressed in points. A reading is complete when at least 90% of the expected answers were obtained; an incomplete reading is flagged and left out of the curves.
Every rate carries a margin of error, based on the number of answers read: over 125 answers, a 32% rate is accurate to within 8 points, 95 times out of 100, and a change smaller than that margin is not a real movement.
4. The figures
| Figure | How it is computed |
|---|---|
| Citation rate | Share of answers to the industry's questions that cite the brand: out of 100 answers, how many cite it. Every answer counts once. |
| Average position | The brand's average rank in the answer, when it is cited. |
| Ranking | The brand's place against its competitors, each measured with the same rate on the same answers. |
| Sources | Domains and pages the AI engines cite, including pages covering the industry without naming the brand: the sources to win. |
| Margin of error | How far the rate read may sit from the true rate, because answers vary: the more answers, the narrower it is. It applies to the citation rate, where every answer counts once. |
| Reputation | Read separately, on the questions that name the brand: what AI engines say about it, and in what tone (positive, neutral, negative). It never counts toward the rate. |
5. The limits
- A measurement is not a guarantee: Hypervisible does not promise any visibility result.
- Models change; their availability and stability are up to their makers. An unavailable engine is flagged and the reading continues on the others.
- Claude and Copilot are not among the five AI engines tracked.
- Customer data belongs to the customer: it is stored in Europe, never sold or pooled into a public dataset.
This method is also published as a citable note with a DOI: 10.5281/zenodo.23740428 (Zenodo, October 28, 2026, in French).
The free analysis applies this method to your brand: 10 questions, five AI engines, 50 answers read in 5 to 10 minutes. Who makes Hypervisible: see About.