This page is the small print behind the front page: where each number comes from, how many properties it is measured over, and what we cannot measure yet.
Data of 15 September 2026. We repeat it every month.
The figures on this page do not come from a report we read. We measured them ourselves.
The denominators: 334 properties open in the Canary Islands tourism register, 272 with a website, 237 whose website served the assistant a page, and 104 of those that also show the guest a nightly rate. A site that was down or throttling us does not enter the figure it could not be measured for. Every number on this page carries its own.
On 15 September 2026 we read each of those sites twice: once the way an assistant reads it, without JavaScript and with the ChatGPT and Claude agents declared, and once the way a guest reads it, in a full browser. The difference between the two reads is what the assistant does not see.
Of the 104 that show a rate to the guest, 40 do not show it to the assistant. 12 of the 272 open the page to a browser and refuse it to the assistant, from the same client and in the same minute. If you are one of those twelve, nothing else on this list matters to you. Four of the 237 we could read publish their rate in a format a machine reads as data. Four. It is the cheapest advantage on this island and almost nobody has taken it.
We repeat it every month. It is the only scan like it in the Canaries: no public dataset covers a single island. If your hotel is in Gran Canaria, we have already looked at it. We do not publish names: yours comes in your report.
In 155 of 237 Gran Canaria properties the assistant finds no rate at all, so it uses whatever it can find insteadVerimundo, 15 September 2026, read the way an assistant reads
79% of hotel links in Google AI Mode go to your Google listing; only 16% go to your websiteNicolas Sitter, nicolassitter.com, 4,000 queries, February 2026
132 of 237 Gran Canaria properties give the assistant no structured description of the hotel, so when it is asked about them it writes its ownVerimundo, 15 September 2026, read the way an assistant reads them
148 of 272 Gran Canaria property websites have a German versionVerimundo, 15 September 2026
Since 31 August 2026, 99.96% of AI Mode citations sit inside a google.com/searchviewer link that hides the source site: only whether your hotel appears can be measured now, not which site it read.
If your site does not show the price, it takes it from wherever it can. ChatGPT sends roughly nine in ten hotel links to the hotel's own site (91.1% on GPT 5.2 across 19,579 runs, December 2025 to January 2026), but it reconstructs the price: 98.8% of price questions trigger a live web search, OTAs are 46% of what it fetches but only 11% of what it cites, and Reddit is cited every time it is retrieved (June 2026). We measure the gap on your own hotel. Gemini does not read your website: it answers from your Google listing. Since 8 September 2026 it answers hotel recommendation questions from the Google Hotels tool with no publisher citations: 408 of 408 questions. Market figures: Nicolas Sitter, nicolassitter.com. A template llms.txt changes nothing.
Market figures: Nicolas Sitter, nicolassitter.com. ChatGPT, December 2025 to January 2026. Gemini, 8 September 2026. AI Mode, February 2026.
What we do not know, and neither do you: how many of those nights start with an assistant today. Your Analytics does not see all of it. Of 35 visitors who said an AI sent them, 24 left no measurable signal at all (Nicolas Sitter, nicolassitter.com, 1 July to 1 September 2026). That is the first figure we measure, and it is why the report is free.
Why the rate: ChatGPT does not read your availability, it reconstructs it (Nicolas Sitter, nicolassitter.com, June 2026). We measure how far off it is on your own hotel.
And we practise it here: this site is readable without JavaScript, with prices as text, a consistent Google listing and a machine-readable index at /llms.txt.
The machine-readable summary, with every figure and its source, is at /llms.txt.