Research
Who AI assistants name for motor and travel insurance in Europe
When someone asks an AI assistant where to buy motor or travel insurance, which brands does it name? We asked OpenAI, Gemini and Claude the same unbranded questions in the local language of 27 markets and counted the answers.
Compiled by: Desislava Tsvetkova, Boleron EU
- Markets
- 27
- Valid answers
- 2,768
- Markets led by a broker or comparison site
- MTPL9of 27Casco3of 27Travel2of 27
Ownership
Who owns the answer
For each market and product, the brand named most often, and how the whole answer splits between insurers and the businesses that sell or compare their products.
- Insurer
- Broker, comparison site or bank
- Assistance provider, motoring club or other
Select a result to open that market in the explorer below.
Explorer
Explore a market
The ten most-named brands for one market and product, and how each assistant differs.
Austria – MTPL
Share of all brand mentions. A brand counts at most once per answer.
- 1durchblicker.at9.8%
- 2Generali7.0%
- 3CHECK246.6%
- 4Zurich6.3%
- 5UNIQA6.3%
- 6Wiener Städtische6.1%
- 7Allianz6.0%
- 8muki Versicherungsverein5.9%
- 9smile.direct5.5%
- 10VAV Versicherung5.5%
| By assistant | OpenAI | Gemini | Claude |
|---|---|---|---|
| durchblicker.at | 10.3% | 6.9% | 12.2% |
| Generali | 6.4% | 8.3% | 6.1% |
| CHECK24 | 5.1% | 4.9% | 9.8% |
| Zurich | 7.7% | 7.6% | 3.7% |
| UNIQA | 6.4% | 7.6% | 4.9% |
| Wiener Städtische | 5.1% | 6.9% | 6.1% |
| Allianz | 6.4% | 5.6% | 6.1% |
| muki Versicherungsverein | 7.7% | 2.8% | 7.3% |
| smile.direct | 5.1% | 4.2% | 7.3% |
| VAV Versicherung | 6.4% | 7.6% | 2.4% |
Who gets named
- Insurer71.9%
- Broker, comparison site or bank20.2%
- Assistance provider, motoring club or other7.9%
Named most by each assistant
- OpenAIdurchblicker.at 10.3%
- GeminiGenerali 8.3%
- Claudedurchblicker.at 12.2%
- Valid answers
- 34
- Backed by live web search
- 97.1%
- Catalogue brands named
- 27 of 34
Example question: Welche ist die beste Kfz-Haftpflichtversicherung in Österreich?
- Insurer
- Broker, comparison site or bank
- Assistance provider, motoring club or other
Method
How it was measured
- Questions
- Six unbranded discovery questions per product and market, written in the local language, asked with live web search enabled and retrieval located in the capital city. No system prompt.
- Sample
- Each question was asked twice to each of three assistants: 36 answers per market and product, before failed or excluded calls.
- Detection
- Brands are found with a fixed per-market catalogue of names, aliases and domains. A brand counts at most once per answer. Rank is the order of first mention. Brands outside the catalogue are not counted.
- Aggregation
- Metrics are computed per assistant and then averaged with equal weight, so one assistant's response volume cannot dominate. Failed calls are excluded from denominators.
- Assistants
- gemini-3.6-flashgpt-5.6claude-sonnet-5
Limitations
What this does not say
- 01Visibility in an assistant's answer is not a measure of product quality, price or an insurer's financial strength.
- 02With two repeats per question, differences of a few points between neighbouring brands are within run-to-run noise.
- 03Only brands in the per-market catalogue are counted. A brand missing from the catalogue is invisible here.
- 04Assistants answer differently over time and by user. This is one run on the date shown, not a standing ranking.
- 05Grounding varies: some answers were written from the model's memory without live web search. The grounded share is shown for each market.
- 06In the 24 September run, Claude answers were collected under two settings: 118 with defaults, 458 capped at 10 web searches with low reasoning effort. OpenAI figures use gpt-5.6 only, and 143 answers from gpt-5.4-mini were excluded.