Insights
Which decisions shape the measurement result, which assumptions hold up under testing, and what can be observed in AI recommendations.
Share of the five most-mentioned insurers, average of ChatGPT, Gemini, Copilot
Study
2 November 2026
Study of 11,566 AI answers: search terms barely change which providers AI mentions, conditions such as a price limit markedly do. Part 1: car insurance.
When a question states conditions such as a price limit or cover, the providers AI systems mention change markedly. Search terms instead of full sentences change little.
Measurement Practice
4 October 2026
A Swiss online shop appeared in 18% of AI answers to four purchase questions and in 3% for eight others, mainly for questions about the best shop, a comparison or returns. We formed the groups after the measurement.
Measurement Methodology
4 October 2026
AI answers vary from query to query and differ considerably between systems. A score without a sample size and an interval is a snapshot. With our own data, including a mistake of our own.
How AI Searches
4 October 2026
AI systems break a question down into their own sub-searches. Pages that rank well there are cited more often, so SEO counts. Whether a page can be read and whether a provider is chosen needs a separate check.
Measurement Methodology
12 September 2026 · updated 4 October 2026
A win rate can be measured cleanly and still be commercially irrelevant. What matters is which decision intents underlie the measurement and where they come from.
Technical Readiness
12 September 2026 · updated 4 October 2026
Client-side rendering is not inherently bad. It becomes a problem when core content only exists after JavaScript has executed.
Further analyses will follow from ongoing measurements and our research track.