Verimundo measures what AI assistants tell guests about a hotel, and whether the hotel’s own site is readable to them. Independent hotels, Gran Canaria. Free, and nothing follows from it unless you ask.
Hotel Ejemplo Playa
Playa del Inglés, Gran Canaria · 120 rooms · 4* adults-only
Illustrative example · fictional data
The assistants answer questions about your hotel without reading your hotel.
- 1 When a guest asks an assistant to find or compare a hotel here, the answer rarely points at your website. It points at Booking.com, at Expedia, or at another hotel.
- 2 What the assistants do say about you is partly wrong: the price, the pool and the cancellation policy were all stated incorrectly on the day we asked.
One cause behind both: the pages an assistant can read do not contain your rooms, your rates or your policies, so it answers from whatever else it can find, which is usually an OTA. The rest of this page is the evidence, in the order we found it.
What we asked,
and what came back
Five shapes of question a guest actually uses before booking: one with no hotel in mind, one filtered by price, one comparison, one price, one channel. They are fixed, so the same five can be re-run later and compared. Each went to four assistants, so “2 of 4” means two of the four named you.
| Question | Assistants that named you | Where the link went |
|---|---|---|
| Best adults-only hotel in Playa del Inglés, one week in January | 2 of 4 | booking.com (2) |
| Hotel near the beach with breakfast for under €200 a night | 0 of 4 | — |
| Compare Hotel Ejemplo Playa with similar hotels | 4 of 4 | booking.com (2), expedia.es (1), own site (1) |
| How much is a double room in January? | 4 of 4 | tripadvisor.es (2), booking.com (1), none (1) |
| Where is it cheapest to book: website, Booking.com or Expedia? | 4 of 4 | booking.com (3), own site (1) |
Named when the guest already knows you. Absent when they do not.
Five questions to four assistants is 20 answers, 14 of which named your hotel. The three figures below are that table counted three ways.
Where the answers
differ from your site
Things an assistant told a guest that your own website contradicts. A wrong price loses the booking, a wrong policy produces a complaint at the desk, and severity is our judgement of which of the two is likely. Anything you publish nowhere is never counted as an error.
| Assistant | What it told the guest | What your website says | Severity |
|---|---|---|---|
| ChatGPT | “Double from €165 in January” | From €212, website rate January 2027 | high |
| Perplexity | “Outdoor pool, not heated” | Heated all year to 28 °C | high |
| Gemini | “Free cancellation up to 24 h” | 72 h on the flexible rate; the website rate is non-refundable | high |
Question 5 in full. Not a complaint about Perplexity: it is an accurate description of your website, and it is the reason the answer sends the guest elsewhere.
“The cheapest option is usually Booking.com, which offers free cancellation; the hotel’s website does not show prices without starting a booking.”
Why this
happens
An assistant does not open your site the way a guest does. It fetches the page, never runs the booking widget, and keeps what is left as plain text. Five things decide whether that read succeeds. We ran them on your site and on the two hotels the assistants recommended in your place.
| What we checked | You | What we found on your site |
|---|---|---|
| Can an assistant reach your website at all? | fail | Your security layer turns two of the four assistants away while serving the same page to a browser. robots.txt fine; CDN returns 403 to GPTBot and ClaudeBot. |
| Can it read your rooms and your rates? | fail | Rooms and rates load after the page does, so those pages arrive empty. Nothing without JavaScript. |
| Is your price written as text a machine can copy? | fail | Price only inside the booking engine, policy only inside a PDF. This produced the Perplexity answer above. No rate as text. |
| Is your hotel described in the format assistants look for? | fail | Your site declares a company, not a hotel: no rooms, no board, no prices. schema.org Organization only. |
| Is there a plain file telling assistants what you offer? | fail | Not present. Worth an hour, not a project: no measured evidence assistants use it yet. llms.txt returns 404. |
Checks 1, 2 and 3 are doing the damage and none of them is a redesign. Sol Dorado is not better marketed than you. It is more readable than you.
Next step
This is one run, on one day, on five questions: enough to show a problem exists, not enough to size it. The detailed review turns it into a list you can hand to whoever maintains your website.
Moving 52 commission nights a month to your own site pays for the work. That is about 3% of the nights you currently sell through OTAs.
How that is calculated: €130 average rate, 75% occupancy, 18% commission, 60% of bookings through OTAs, so roughly 1,620 commission nights a month at €23 of commission each. Every one of those is a guess until you correct it on the call. This is not what AI is costing you and it is not the price of the fix: it is the bar the work has to clear. If 52 nights is more than you sell on commission, do not buy this.
- 25 guest questions instead of 5, five runs per assistant instead of one, so the result is a share and not a lucky answer
- Every claim checked against your own fact sheet and your Analytics
- A fix list ordered by impact, each item costed in hours
Book 20 minutes at verimundo.com, or reply to the email this came with and name a time.
Verimundo · Las Palmas de Gran Canaria · verimundo.com · Public assistant answers on the date shown. Results vary between questions and between days, which is why this is presented as one run and never as a ranking, and why the detailed review measures shares across five. We name no hotel publicly, including yours.