AI can quote your hotel’s own website and still give guests the wrong answer
Lumina research found a $410 gap between an AI-quoted seaplane fare and a Maldives resort's live website price, traced to a 2015 guest review used as the model's source.
The seaplane fare in a guest review and on the resort’s website
Lumina Frontier (AI-generated illustration)
When we traced the errors in AI-written fact sheets for 31 Maldives resorts, the most common mistake was choosing the wrong source when several disagreed. That accounted for 18 of the 38 errors we found. We deliberately included difficult cases in the check. What interested me as someone who writes about hotels was how easily one of those answers could pass an editor’s simplest test. Open the link and there’s the figure, on the hotel’s own website.
In a separate check, our AI put a seaplane transfer at $1,500. We traced the figure to a guest review from 2015. At the time of our check, the resort’s own page listed $1,910. A traveller comparing two resorts would have been budgeting $410 too little for the journey to one of them. Going straight to the hotel’s website helps, but the date still matters. In another case our fact sheet quoted $488 return. The page gave that fare for travel from January 2023 and gave no current price.
If I ran the hotel, I’d start with the last practical detail we changed. Can reservations show me what a guest will find about it today, including in our old PDFs? And if those pages disagree, who has the authority to settle the answer before an agent sends it out under our name?
What we missed on 119 pages
We wrote on 119 pages that a hotel reseller didn’t tell you which room you were buying. I could see why we’d said it. Google Hotels showed the price and left out the room name. One more click would have answered the question. The reseller named the room on its own site.
We fixed the pages. We also changed the instructions our agents read before they write. If we make a claim about what a seller tells a guest, we have to visit the page a guest would see. “Check your facts” had always sounded clear enough to me. It took those 119 pages to realise it left out where to look.
What the second hotel should inherit
We found a Tuscan resort whose house rules gave the dog fee as €35 per stay and whose FAQ said €35 per night. Suppose the resort confirms the fee is per stay. Someone at the property can now answer the guest and get the FAQ changed. The group marketing team has a different job. It can give every hotel the same instruction for the next time two official pages disagree. Keep both passages, flag the difference and send it to the person at that property who sets the policy. The next hotel is free to charge per night.
I’d try this with whoever runs marketing across the group. Take the last correction reservations sent you. Open the pages a guest could still find and see whether any of them repeat the old answer. Then sit with the marketer at another property and ask them to draft a reply to the same enquiry using their own hotel’s sources. Watch where they have to make a judgement and whether the instructions give them someone to call.
If the two pages disagree, I’d count an unfinished reply as a successful test, provided it lands on the right desk with both passages attached. I’d trust that team with a second hotel.
Marco Di Cesare is Founder & CEO of Lumina Frontier, a Zurich company building an AI marketing team for hotels.
Comments
Comments for this content
0 comments available