Hotels
"Where to stay near Mont-Saint-Michel" is exactly the kind of question people no longer put to Google. We measured what the AIs answer, across 90 real queries.
Report within 48 hours. Secure payment by Stripe.
The study
10 traveller questions — family, budget, romantic weekend, group, guest house, wheelchair access — put to 3 AI assistants, three times each. 90 answers, in French, English and German. We counted every property named.
One property is named in 24% of answers, three times more than the runner-up. In total, seven names share every answer — in a bay that counts dozens of places to stay. The others do not exist for a traveller asking an AI.
« Je cherche un hébergement pour 6 personnes autour du Mont-Saint-Michel en août »
Sources read by the AI
Not one property's own website. The AI reads only platforms and the tourist office.
« Existe-t-il des chambres d'hôtes de caractère dans la baie du Mont-Saint-Michel ? »
Sources read by the AI
On niche questions, individual properties finally surface. That is where it is won.
« Hôtel de charme près du Mont-Saint-Michel pour un week-end en amoureux ? »
Sources read by the AI
Six properties named in a single answer — and the same six come back everywhere.
A traveller is not looking for a hotel, they are looking for a stay. "With the family", "cheap, under 15 minutes away", "with a good restaurant", "wheelchair accessible". These are questions a booking platform answers poorly and an AI answers very well — by naming two or three properties.
Our measurement surfaces two things.
AIs read platforms and editors, not hotel websites. Booking, the tourist office, TripAdvisor and three travel blogs account for most of the sources consulted. A property's own site only appears once questions get specific.
But they name places, not platforms. The most-cited property in the bay is named twice as often as Booking. Which means the ground can be taken — provided you exist inside what the AIs read.
How we measure
No question contains your name. That is the condition for the result to mean anything: we look at whether the AI mentions you spontaneously, not whether it can describe you when prompted to.
From your business, your website and your territory, we write the questions your buyers really ask — in their words, and in their languages.
Perplexity, ChatGPT and Claude, with web search enabled. Each question is asked three times, because an AI never answers the same way twice: a single measurement proves nothing.
We count your citations, your recommendations, where they sit in the answer, and every domain the AIs went to read to form an opinion.
The assistants we query
Three engines, all with web search enabled — under the real conditions in which your buyers use them.
Sonar
Built-in web search. The most source-driven engine of the three.
GPT, consumer model
The model actually served inside ChatGPT, not an API variant.
Sonnet 4.5
Native web search, long and highly structured answers.
A report you read in fifteen minutes and turn into decisions.
Out of 360 answers, how often your site is a source the AI actually read. Broken down per assistant.
How often your brand is named in the answer, and in which position — being cited first is not worth the same as being cited last.
A ranking of the brands AIs recommend instead of you. Not the ones you think you have: the ones the data points to.
The domains cited most often in your market — forums, comparison sites, trade media. That is where your presence is decided.
The question-by-question detail, so you can see exactly which topics you do not exist on.
The three or four levers your data points to: which domains to get cited on, which topics you are absent from, what is holding your site back. Derived from the measurement, not from general principles.
One audit, one price
€149excl. VAT
One-off payment. No subscription, no commitment.
Five fields. They are what we use to write your 40 questions — the more precise you are, the more accurate the audit.