Ask five AI tools the same simple travel question, and you will get five different answers. That is not a hunch: it is one of the findings in a new study by Skift Research and McKinsey, Winning Hearts in an Age of Infinite Travel Choices, published this week.
I read it with two hats on. As a traveller who plans almost everything herself, and as someone who has run a small hotel in the Cinque Terre since 2008, I know exactly what goes into the descriptions these tools are reading. Here is what the research found, why it happens, and how I would use AI to plan a trip without letting it plan the trip for me.
Five tools, one question, no agreement
McKinsey’s researchers asked three general AI assistants and the AI tools of two online travel agencies the same question: the top three kid-friendly hotels in Chicago with river views, for the Labor Day weekend. Not a single hotel appeared in all five answers. Most appeared in only one.
To be fair, this was one prompt, run as an illustration rather than a scientific test, and the wider survey behind the report covers just over a thousand American travellers. But anyone who has tried it at home will recognise the pattern. Each tool has its own favourite, and each presents it with exactly the same confidence.
The practical consequence is a little ironic. AI was supposed to save us research time, yet nearly 60% of the travellers surveyed say that although it gives them answers faster, they end up spending more time researching overall. Once you realise every tool has a different “best”, you start asking all of them.
I tried it on my own stretch of coast
Curious, I ran my own version of the test at home. I asked ChatGPT, Gemini and Claude for a place to stay with a garage and a really good breakfast between the Gulf of Poets and Sestri Levante, the stretch of Ligurian coast I know best. Between them, they named twelve places. Not one appeared in all three answers, and only one appeared in two (in full disclosure, it was my own hotel in Levanto, so I will say no more about it). Everything else came up just once.
Then I checked every single recommendation, and that is where it got interesting. ChatGPT and Claude explained their choices and linked their sources. Gemini offered three beautifully written suggestions, complete with homemade cakes and focaccia served in the garden, and not one link. The first two do not exist: there is no “Villa Camilla Boutique Hotel” in Sestri Levante and no “La Dimora del Centro” in Levanto. The third is real, but it has paid outdoor parking rather than a garage, and it does not serve breakfast at all. Three recommendations, three misses, all delivered with complete confidence.
Why “family-friendly” gets you nowhere
The most useful detail in the report is about which hotels kept coming up. The ones that surfaced most consistently were those whose content described specific, documented features, such as a suite designed for families with children. Hotels that relied on generic phrases like “family-friendly” tended to slip away.
This makes complete sense from behind the reception desk. An AI tool builds its answer by matching what you asked for with facts it can find repeated across several sources. A vague adjective matches everyone, so it tells the machine nothing. “Connecting rooms, a cot on request, a fenced pool, 200 metres to the beach” is something it can find, compare and quote back to you.
The lesson for travellers mirrors the lesson for hotels. The AI’s recommendation is only as good as the facts the hotel has bothered to publish. A lovely, quirky, well-run place with a lazy online description may simply never make the list. And a big property with a very thorough listing may make every list without being right for you at all.
What AI cannot see: the Spanish castles problem
My favourite example in the report comes from Turespaña, Spain’s national tourism board. Its director general explains that when you ask AI models where to discover castles in Europe, Spain often does not appear, even though it has some of the highest numbers of castles on the continent.
The reason is revealing. The models lean heavily on the inventories of online travel agencies and tour operators. German castles are sold online as packaged tours and tickets; many Spanish ones are not. So, as far as the machine is concerned, they barely exist.
For an independent traveller, this is actually good news, because it tells you where the gaps are. The village festival nobody sells tickets for, the hill path with no guided tour, the family-run guesthouse with no marketing budget: these are exactly the things AI is most likely to miss and the things I travel for. If you want them, you will still find them the old way, through books, local tourist offices, other travellers and people who live there. I wrote about why that matters in How Not to Love a Place to Death: the places that only appear where everyone else looks are the ones that suffer most.
Most of us want ten good options, not one answer
The number that struck me most is this. Travellers were asked to imagine three platforms with identical prices and availability: one showing 500 hotels, flights and experiences with filters, one showing ten recommended options based on their budget and preferences, and one giving a single AI-generated trip plan. Sixty-three per cent chose the ten curated options. Thirty per cent wanted the full catalogue. Only 7% wanted the single AI plan.
In other words, more than nine in ten of us want to choose for ourselves. That is also my experience. I like help narrowing the field; I do not want someone else, human or machine, making the final call on where I sleep. As the Columbia professor Sheena Iyengar puts it in the report, travellers need to feel that they made the choice.
How I would actually use AI to plan a trip
Here is the method I would suggest, and it fits neatly with how I describe planning an independent trip you can improvise.
Ask more than one tool. Put the same question to two or three assistants. Where their answers overlap, you are probably looking at a genuinely strong option. Where they differ, you have a wider shortlist.
Ask for reasons and sources. Do not accept “this hotel is great for families”. Ask what, specifically, makes it suitable, and where that information comes from. If the answer stays vague, treat the recommendation as vague too. And be most wary of an answer with no sources at all: in my own test, that is exactly where the invented hotels came from.
Build your own shortlist of ten. Use AI to go from hundreds of options to a handful, then take over. Keep your list somewhere you control (I still use a simple spreadsheet), so you don’t have to start again every time you switch apps.
Check the facts at the source. Rates, availability, opening days and policies change constantly, and AI answers can be out of date. Before you commit, check the hotel’s own website, the operator’s page or the official tourist office. If something matters to you- a ground-floor room, a late check-in, a quiet street- write to the hotel and ask. Small hotels answer, and the reply tells you a lot about the place.
Read reviews for specifics, not stars. A review that mentions the cot, the stairs or the noise from the square is worth ten that just say “amazing”.
Book the last step yourself. Only 16% of the travellers surveyed have used an AI tool to book anything, and three in four prefer to make the final booking themselves. I am firmly in that camp. For how I choose between booking direct and through a platform, see my guide to hotel bookings from both sides of the desk, and for the planning apps I actually keep on my phone, my list of the best travel apps for independent travellers.
A word from the other side of the desk
One more finding worth knowing: the report says 51% of travellers have changed a hotel booking after seeing a better option on social media, and many keep checking prices after they book. Free cancellation makes that easy, and I understand the temptation.
As a hotelier, I would only ask one thing. If you find a better price elsewhere for the same hotel, write to the hotel before you cancel. Often, we can match it or add something the platform cannot, and you keep the room you chose in the first place for good reasons.
AI is a remarkable research assistant. It is not yet a travel companion. Use it to widen, then narrow, your options; keep the final choice for yourself; and leave some room on the map for places no algorithm has found yet.
Further reading
- McKinsey and Skift Research, Winning Hearts in an Age of Infinite Travel Choices: https://www.mckinsey.com/industries/travel/our-insights/winning-hearts-in-an-age-of-infinite-travel-choices
- Skift, coverage of the report at Skift Global Forum 2026: https://skift.com/2026/09/23/what-ihg-priceline-and-mckinsey-say-brands-are-missing/
- Independent Travel: How to Plan a Trip You Can Improvise: https://silvias-trips.com/independent-travel/
- Hotel bookings, from both sides of the desk: https://silvias-trips.com/hotel-bookings/
- The Best Travel Apps for Independent Solo Travellers: https://silvias-trips.com/best-travel-apps/
FAQ
Each tool draws on a different mix of sources, such as booking platforms, hotel websites, travel media, reviews and maps, and weighs them differently. In a McKinsey test using five AI tools, no hotel appeared in all five answers, and most appeared only once.
Use them as a starting point, not a verdict. Ask the tool why it recommends a hotel and where the information comes from, compare two or three tools, and check prices, availability and specific features on the hotel’s own website before booking.
AI models rely heavily on what travel agencies and tour operators sell online. Places and experiences that are not packaged as bookable products, or written about by authoritative sources, are often left out, as Spain’s tourism board found with its castles.
Yes. When I asked three AI assistants for a hotel with a garage and a good breakfast on the Italian Riviera, the one that cited no sources suggested two hotels that do not exist and a third without the services requested. Prefer answers with verifiable sources, and always confirm details on the hotel’s own website.
Most travellers do not yet. According to the Skift and McKinsey research, only 16% have used an AI tool to book anything travel-related, and three in four prefer to make the final booking themselves. Checking details and booking directly gives you more control if something goes wrong.
Use it to generate ideas and narrow hundreds of options to a shortlist, then take over: build your own list, check facts at the source, read reviews for specific details, contact small hotels directly, and leave room for places that no AI tool has picked up.
One Comment