AI isn’t personalising travel. It’s personalising the sales pitch. Ask ChatGPT, Gemini or Claude for a trip idea as a family, a backpacker or a couple with money to burn, and the words change to suit you. The places mostly don’t. That isn’t a technology failure. It’s a warning to every destination that has assumed AI would do its discovery work for free.
If you run a smaller destination, a regional tourist board or a hotel outside the obvious cities, this matters more than any chatbot launch this year. AI has become a very polite way of sending people to the same places they were already going.
What the research actually found
Skift Research has just published a report called The Mystery of the Disappearing Destination. The team ran repeated prompts through ChatGPT, Gemini and Claude using different traveller personas, with different ages, interests, travel companions and budgets.
The models did adapt. Families were told about family-friendly activities. People on tight budgets were pitched low-cost flights. But the destinations barely moved.
The numbers are striking. The top five destinations took 38% of all AI recommendations. Those same five places take about 40% of real global inbound arrivals (2024 figures). At the top ten level, AI came in at 48% against 56% of actual arrivals.
I borrowed one of those numbers last week to argue that interchangeable hotel brands are the real loyalty problem. For destinations, it’s the whole story. AI recommendations are almost as concentrated as real-world travel. It isn’t creating new demand patterns. It’s copying the old ones and wrapping them in friendlier language.
The shortlist is the real threat
Here’s the bit that should worry destination marketers most. A Google results page gives you hundreds of links. An AI answer gives you a handful of options with reasons attached. And travellers like it that way.
In a Skift Research and McKinsey survey of 1,015 US travellers, 63% said they’d rather have a curated shortlist of ten options. Only 30% wanted a big catalogue of 500 options, and just 7% wanted a single AI-built itinerary they could edit.
So the game is no longer “rank somewhere on page one”. It’s “be one of the ten names the machine says out loud”. If you’re eleventh, you don’t exist. I made the same point about hotels last week: AI isn’t snubbing independent hotels, it just hasn’t heard of them. Destinations are now in exactly the same position.
Tourism Ireland shows it can be shifted
The good news is that this isn’t fixed. Tourism Ireland told Skift’s Transatlantic Summit in Dublin that information it supplies now shapes close to one in five AI-generated answers about Ireland. Its own research found the share of travellers naming AI as their main source of travel inspiration doubled in the past year.
Chief executive Alice Mansergh said that, left alone, AI repeats the same familiar recommendations. So Tourism Ireland made its digital content readable by generative AI, started tracking the results, and began pushing smaller experiences, such as sheepdog demonstrations in Sligo and traditional boat building on the Aran Islands, to spread visitors around the island.
Her advice was simple: choose what you want to be famous for, then build “a bit of a surround sound” of other voices saying it, because the models also pull from Reddit, YouTube and Tripadvisor.
One fair caveat: Tourism Ireland hasn’t published how it measures that one-in-five figure, and Skift says it has asked. I’d like to see the method. But the direction is right, and almost nobody else has even put a number on it.
Stop blaming the algorithm
I hear a lot of destination people say AI is “biased towards the big names”. That’s the comfortable version. The less comfortable version is that the big names have years of articles, reviews, videos, forum threads and structured listings behind them, and most smaller places have a brochure PDF and an events page that hasn’t been updated since spring.
AI can only recommend what it can find, read and trust. Skift Research’s own conclusion is that destinations should treat AI visibility as a distribution problem, not a branding one: better structured local content, and working with the operators and platforms that feed information into these systems. I agree, and I’d go further. A beautiful brand campaign that AI can’t read is now a very expensive mood board.
We’ve seen the same pattern across travel. Airlines are named in AI answers but don’t get the click. Cruise lines are moving inside ChatGPT and Claude because the website is no longer the front door. Destinations are just the next layer up.
What I’d do this quarter
- Audit the answers. Ask the main AI tools the questions your visitors actually ask, across different personas, and record whether you appear.
- Pick three things to be famous for. Not twenty. Specific, real experiences that nobody else can claim.
- Write for machines and people. Clear pages with facts, locations, seasons and prices, not just adjectives.
- Build the surround sound. Get your operators, creators and partners talking about the same things on the platforms AI reads.
- Don’t flood it with slop. Mansergh warned about consumer distaste for “AI slop”. Thin AI-written pages won’t earn trust from people or models.
So here’s the debate I want to have. If AI keeps sending travellers to the same top ten, is that the model’s fault, or is it the fault of every destination that never gave it a reason to say anything else?
If you want more takes like this on AI, search and travel marketing, subscribe to my LinkedIn newsletter.