Method checked in November 2026
How we measure.
Ask the same question twice and you often get two different lists. So we count shares across many questions and runs, never a single rank.
In short
We measure whether AI assistants recommend a brand. To do so, we put real buying questions in German and French, depending on the category also in English, to ChatGPT, Google AI Overviews, Gemini, Claude and Perplexity and count in what share of answers the brand appears, who is named instead and which websites the answers cite as sources. We query Gemini with Google Search and Claude with web search through the official Google and Anthropic interfaces, ChatGPT, Perplexity and the Google results page through the data service Apify. We never use a login or a personal account.
- Buying questions in two to three languages
- 5 AI systems, Copilot by hand
- Shares across many answers
- Sources for every answer
Foundation
Built on research, measured in Switzerland.
Our method is based on research into Generative Engine Optimization (GEO). The foundational study by researchers including Princeton University shows that content with source citations, quotations and figures becomes up to 40% more visible in AI answers.
We have developed this mechanism for Swiss brands. So far we have measured 1,024 Swiss brands from seven categories, among them cosmetics, jewellery, watches and drinks (as of 16 November 2026), in German, French and English.
Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan, Deshpande: “GEO: Generative Engine Optimization”. KDD 2024. arXiv:2311.09735
1, The questions
Questions the way customers ask them.
«Which Swiss face cream is good for sensitive skin?» instead of «face cream Switzerland». In German and French, depending on the category also in English: for the best brand, for a solution to a need, for a gift or a place to buy, as people in the category ask.
| Questions | Systems | Runs | |
|---|---|---|---|
| Quick check | none | none | 22 technical checkpoints |
| AI analysis, free | 10 to 40, depending on the category | up to 5 | 1, checked by hand, plus the 22 checkpoints |
| Audit | 40 to 60 | 5, plus Copilot by hand | 3 |
| Ongoing support | 40 to 60 | 5, plus Copilot by hand | 3, every month; plus a weekly check with 10 core questions |
2, The systems
Every access labelled openly.
- ChatGPT
OpenAI search model, through the data service Apify.
Analysis yesAudit yes - Google AI Overviews
Google.ch, in the language of the question, through the data service Apify. They do not appear for every question; we also count how often.
Analysis yesAudit yes - Gemini
Google's interface with Google Search, not the Gemini app. We record which brands and sources are named.
Analysis yesAudit yes - Claude
Anthropic's interface with web search. We record which brands and sources are named.
Analysis yesAudit yes - Perplexity
Sonar, with sources, through the data service Apify.
Analysis yesAudit yes - Microsoft Copilot
No official interface with web search. Manual spot checks in audit and support.
Analysis noAudit by hand - Google AI Mode
Available in Switzerland, but currently returning no answers through our data service (as of October 2026). We check again in every audit.
Analysis noAudit once available
We never use a login or a personal account. That way we see answers without a personal history.
3, The metrics
A score from 0 to 100.
Share of all AI answers that name your brand (share of answer)
60%Share of answers that give your website as a source
25%Share of questions for which you are among the top ten results
15%The quick check: access, product page, trust, Switzerland, technical. A separate value from 0 to 100, not part of the score. Without an AI measurement the report shows only this value
0%More important than the score are the individual findings: share of voice, position, who is named instead and which sources the answers come from.
4, The sources
The source map.
AI systems with web search rely on marketplaces, comparison articles, tests, review portals and your own website. We count which pages are cited most often and check whether your brand appears there.
We do not need access to your shop and work only with public pages. No ranking guarantees, no fake reviews, no health claims. Questions to hello@shortlisted.ch.
5, Sample
We checked 60 Swiss shops.
Technically, most are readable. What is missing is what AI needs for a recommendation.
Median technical readability.
Your brand
This is exactly how we measure your brand.
10 to 40 buying questions depending on the category, five AI systems, within 48 hours. Free.