top of page
HENSEI-LOGO-V2.png

AI and Travel: How Chatbots Changed Our Vacations, Without Managing to Sell Them to Us

Writer: Stéphane Guy
Stéphane Guy
5 days ago
11 min read

We all know someone who asked ChatGPT to build a road trip in ten seconds instead of burning three evenings on forums. That's no longer a geek anecdote. In the US, 56% of leisure travelers used AI for at least one trip over the past year, up from 43% in late 2025 and more than double the 2024 share. France follows the same curve: 40% of travelers in 2026, against 19% a year earlier and just 9% in 2024. But here's what few articles say plainly: the shift stops dead at the checkout. At Booking Holdings, one of the world's largest accommodation distributors, traffic from large language models still represents less than 1% of room nights. AI changed how we dream up a trip. So far, it has barely changed how we pay for one.


Une femme en voyage avec sa valise
Photo by Ross Sneddon on Unsplash

In short

  • Adoption is exploding: 62% of travelers worldwide are now familiar with AI trip-planning tools (Skift Research), and heavy use among US travelers more than doubled in a single year. In France, usage has more than quadrupled in two years.

  • Booking is holding out: AI-referred traffic is below 1% of Booking's room nights, and only 8% of travelers say they're comfortable booking through an AI platform, according to Expedia Group.

  • Hallucinations have a real cost: tourists were left stranded at altitude in Peru and on a Japanese mountain after following itineraries a chatbot invented from scratch, and park staff in the Pyrenees are logging similar cases.

  • The industry is moving at two speeds: only 22% of destination marketing organizations offer an AI tool, against 54% of travelers who have already used one.

  • The real risk may not be the one you think: by recommending the same "hidden gems" to millions of people, AI could manufacture the overtourism it claims to avoid.


What has really changed in our travel habits


The shift happened fast, almost too fast for the industry to organize itself. Skift Research's State of Travel 2026, now in its fifth edition, finds that 62% of travelers worldwide are familiar with AI trip-planning tools. That's no fad running out of steam. Phocuswright's tracking shows usage more than doubling since 2024, and Skift's US tracker saw the share of travelers using AI tools extensively jump 124% between 2024 and 2025. 


The most revealing detail, in fact, has nothing to do with leisure travel but rather with business travel. In the United States, 86% of business travelers report using AI to plan their trips, compared to 55% of leisure travelers. Skift explains this 31-point gap quite simply: a consultant who takes the same flight every week to visit the same client has a route that’s repetitive enough for AI to optimize it without any surprises. Vacations, on the other hand, are all about the unexpected, something we don’t always want an algorithm to handle.


This shift also changes how destinations get discovered. The classic funnel (search, compare, buy) is turning into something closer to an orbit: travelers circle back, re-compare, change their minds, sometimes after they've already booked. In the joint Skift Research–McKinsey report Winning Hearts in an Age of Infinite Travel Choices, 71% of travelers are comfortable letting AI generate initial ideas and 67% are comfortable letting it narrow and compare options, yet only 16% have used an AI tool to complete a booking. Meanwhile, hotels face a visibility cliff. One estimate cited at Skift's Data + AI Summit puts the share of hotels that surface in AI answers at just 6%. Most of the market is invisible at the traveler's very first question. This is the move from SEO to GEO applied to hospitality, with the same winners and the same also-rans as in any other sector that missed the turn.


The great divide: we plan with AI, we still don't book with it


This is, in my view, the most under-exploited piece of information right now, and the one worth dwelling on instead of repeating the same list of tools everyone else does.


Expedia Group's AI Trust Gap study, commissioned by Expedia and run by YouGov among more than 5,700 adults in the US, UK and India, draws a very clean border. 53% of travelers are comfortable letting AI suggest options, and 40% use it to build an itinerary. But when it's time to pull out the credit card, 68% prefer a trusted brand and 66% won't trust an AI assistant to buy or book on their behalf. The result: only 8% say they're comfortable booking through an AI platform. Expedia's chief AI and data officer, Xavi Amatriain, sums it up in a line that applies to far more than travel: "Travelers don't have a technology problem with AI. They have a trust problem."


The same gap shows up on the distributor side. On Booking Holdings' second-quarter 2026 call, CFO Ewout Steenbergen said LLM-referred traffic, paid and unpaid, remains below 1% of room nights, and that the figure "hasn't moved so much recently." Skift found that only 2% of US travelers were willing to give an AI tool full autonomy to make bookings, while 80% of senior travel executives plan to deploy agentic AI at scale within three to five years.


This isn't an ergonomics problem a better interface will fix. It's a question of accountability: if an AI agent books the wrong room, cancels the wrong flight or pays twice for the same hotel, who answers for the mistake? The traveler, the platform, the model's publisher? Skift's own reporting asks who foots the bill when an agent gets it wrong, and finds no clear answer, and the law is still catching up (we break down who is liable when AI makes a mistake separately). Until that question has a clear answer, no interface, however polished, will unlock the door.


Une femme sur son ordinateur portable
Photo by Luke Southern on Unsplash

When AI invents your next destination


It's the story that did the rounds of travel desks last year, and it deserves to be told in detail because it shows exactly how the mechanism works. Two tourists were preparing to hike alone, without a guide, near Mollepata in Peru, toward a place their chatbot had dubbed the "Sacred Canyon of Humantay". The catch: that canyon doesn't exist. Miguel Angel Gongora Meza, a local guide and co-founder of Evolution Treks Peru, intercepted them in time. He told the BBC the name was stitched together from two unrelated places, and that the coordinates the AI supplied could have led them into a potentially dangerous high-altitude zone. One of the travelers had already paid roughly $160 to reach a rural road leading... nowhere. "This sort of misinformation is perilous in Peru," he said, in a country where altitude, weather and trail access have to be planned for seriously.


The same BBC report features Dana Yao, who set off with her husband to climb Mount Misen, on Japan's island of Itsukushima, following ChatGPT's last-ropeway time to the letter. Except the ropeway had already closed, and the couple ended up stranded on the summit. Rayid Ghani, a machine-learning professor at Carnegie Mellon, explains the mechanism without sugarcoating it: a chatbot can't tell travel advice from a driving route or a recipe. It strings together statistically plausible words, not verified facts, and the more fluent and precise the phrasing, the more credible it sounds, even when it's entirely wrong. How often does that happen? A Greetwell survey of 1,000 US leisure travelers found that 55% of those who had used AI for travel hit at least one bad recommendation, and about one in six was sent to a place, tour or activity that wasn't real. It's a survey from a travel concierge company, so read it as an indication, not a census.


France isn't spared. Park authorities are beginning to document similar incidents. In the Hautes-Pyrénées this July, two 18-year-old hikers heading for Lac d'Estom, above Cauterets, followed a route prepared with ChatGPT that sent them toward the Wallon-Marcadau refuge, about 20 km from their goal. A park agent caught the error and no rescue was needed, and he told local radio he had met about thirty people lost since the start of summer because of AI-generated or web-sourced maps. This spring, a father and son walked into the Cévennes National Park's visitor center in Florac with a ChatGPT plan for 82 km in two days, including a hammock night above 800 meters in an area where camping is banned; park staff rebuilt the route from scratch. What strikes me in these stories isn't the error itself (a closed site, an underestimated distance) but the assurance with which it's delivered. You instinctively distrust an out-of-date paper map. A text generated with the fluency of a seasoned guide, much less so.


This is exactly how a language model behaves: it predicts plausible text, it doesn't check facts. Travel is just one use case among many where a confident tone guarantees nothing about accuracy, except that here the mistake is sometimes paid for in hours lost at 4,000 meters with no signal.


The new tools redrawing the traveler journey


Beyond generic chatbots pressed into service, a new generation of specialized tools is taking hold. Expedia launched an app inside ChatGPT in October 2025 and added a Claude connector in May 2026, but in both cases you search in the chat and click through to Expedia to complete the booking. In France, SNCF Connect launched its own ChatGPT app in April 2026, with the final transaction staying inside the SNCF Connect ecosystem, an experiment its then-CEO, Anne Pruvot, discussed at VivaTech 2026. Note the pattern: the chatbot does the dreaming, the brand keeps the checkout.


What these tools change in practice isn't the amount of information available (it was already overwhelming before AI) but how we reach it. We no longer compare fifteen browser tabs: we ask a question and get a synthesis. The time savings are real. The flip side is that we also delegate the sorting of information to a system whose reasoning we can't see, a trade-off we explored in the real cognitive cost of outsourcing your thinking.


That delegation creates a striking shift in trust. According to KAYAK's What the Future 2026 report, 41% of Gen Z and millennial travelers now trust an AI recommendation more than people or social media, and 34% say AI feels more accurate and up to date than peer suggestions. There's something dizzying about that figure: the "tip from a friend who's been there," a pillar of tourism forever, is starting to lose ground to an algorithm that has never set foot anywhere.


A two-speed tourism industry


While travelers equipped themselves in a matter of months, the industry is moving far more slowly. In a study by Mindtrip, Sabre, Sojern and MMGY Travel Intelligence, 54% of travelers have already used AI to plan a trip and 75% intend to do so again. In contrast, only 22% of destination marketing organizations (tourist offices, regional boards) offer an AI tool on their website, and just 26% have a documented AI strategy. The gap isn't closing soon: traveler usage has climbed about 70% in a year, while institutions advance in small steps.


That imbalance has a direct side effect on jobs. In RateHawk's global survey of travel agents, 44% say AI could eventually replace them, and online agency lastminute.com is reportedly cutting about a quarter of its workforce, around 400 jobs, by the end of 2026 to restructure around AI. The paradox is that the trust gap partly protects independent agencies: Skift's analysts argue that human travel agents remain relevant 25 years after the invention of the web, since complex bookings benefit from a traveler working alongside an agent. Travelers plan with AI but still mostly book through a brand or a human they trust. The threat doesn't come from a customer deserting the agency for a chatbot. It comes from platforms capturing the value of planning upstream, leaving professionals the less rewarding role of last safety net at the moment of payment. We detail this broader mechanism, which reaches far beyond tourism, in the jobs AI will kill, transform or create.


Le pont d'or au Viêt Nam
Photo by Vivu Vietnam on Unsplash

The trap of the personalized tourist bubble


Here's the angle I find the most underrated of all, and the one where I'd rather take a position than stay purely descriptive. The promise of AI in tourism is personalization: you describe your tastes, it hands you a tailor-made itinerary, far from the crowds. But that promise rests on a fairly simple statistical blind spot. If millions of people make, give or take, the same request ("somewhere authentic, away from tourists, off the beaten path") to models trained on the same corpora and fed the same popularity data, they will statistically converge on the same ten or twenty answers. Off the beaten path becomes the new beaten path, just recommended one notch earlier in a place's tourism life cycle.


The evidence is starting to pile up: a research letter in Current Issues in Tourism, analyzing 50 structured ChatGPT conversations, found a strong tendency to favor iconic landmarks and high-density tourist hubs, with alternatives appearing less often and typically only when explicitly prompted (Mellors, 2025). And the demand is primed: 69% of Gen Z and 66% of millennial travelers told KAYAK they prefer places "off their feeds".


The question is now on the UN's agenda. In his World Tourism Day message, Secretary-General António Guterres warned that, left unchecked, AI could "funnel tourists to the same few places and platforms" (Travel Tomorrow). Will AI become a tool for regulating flows, steering visitors toward less congested alternatives, or an accelerator of personalized bubbles that close each of us inside our own recommendations, at the risk of homogenizing the destinations people actually visit? I lean toward the second, at least as long as models are trained on aggregated data rather than on a genuine diversity of local, up-to-date sources. This convergence mechanism echoes a broader phenomenon: a system trained on existing data structurally tends to amplify the trends it observes rather than correct them.


How to use AI to travel without getting trapped


Nothing here justifies dropping these tools; they save real time on the initial groundwork of a destination. A few reflexes cut the risks markedly:


  • Cross-check sensitive information: transport schedules, trail openings, visa requirements. Verify them against an official source (local tourist office, transport operator, embassy) before you leave.

  • Be wary of place names that are too poetic or too perfectly tailored to your request. A "sacred canyon" or an attraction nobody else mentions online deserves a check via image search or a recognized map.

  • Favor refundable bookings while you're testing an AI-generated itinerary, especially for physically demanding outdoor activities.

  • Call a human for anything touching physical safety: altitude, shifting weather, hiking permits. Local guides and national parks remain the reference on these points.

  • Remember that AI doesn't know your real profile: your fitness level, the toddler in tow, the lack of an off-road vehicle. It answers the question asked, not your whole situation, unless you spell it out.


What I take away from this shift


AI hasn't revolutionized tourism in the sense usually meant: it hasn't replaced the travel agency, it hasn't killed Google search, and it hasn't really changed who pockets the sector's money. What it has done is move the trust cursor upstream. We hand it the dream, the inspiration, the first draft of an itinerary, and when it's time to pay, we go back to what we already know. Earlier waves of tourism disruption (Tripadvisor, comparison sites, social networks) ended up redistributing power in the industry. This one, so far, doesn't: it adds a layer of conversational search on top of a booking system that has barely moved. How long that status quo holds, and whether the generation that already trusts an algorithm more than a friend will keep verifying its sources once it's on the ground, remains to be seen. I have a reasonable doubt about it, and I think it's worth keeping in mind before letting an AI build, alone, your next high-altitude trek.


FAQ


  1. Can AI book a trip for you? 

    Technically, some platforms already allow it: Priceline's Penny can take you from idea to booking inside the conversation, while Expedia's ChatGPT app and Claude connector hand you off to Expedia to complete the purchase. In practice, very few travelers take the leap: only 8% say they're comfortable booking through an AI platform, according to Expedia Group.


  2. Is it safe to use ChatGPT to plan a hike or a trip? 

    Not without verification. Documented incidents in Peru, Japan, the Hautes-Pyrénées and the Cévennes show that chatbots can invent places, underestimate distances or ignore closing times and local camping rules. A general-purpose AI doesn't know real-time trail conditions or your fitness level.


  3. Will AI replace travel agents? 

    Not immediately, and not in the way most people imagine. Routine ticketing is exposed, but the trust gap that blocks direct AI booking currently benefits human professionals at the moment of payment. In RateHawk's survey, 44% of travel agents think AI could eventually replace them, which means a majority don't.


  4. Why do travelers use AI to plan but not to book? 

    Studies converge on the same brakes: a perceived loss of control, worries about payment data, and no clear customer service if something goes wrong. In Expedia's survey, more than half cited loss of control and data privacy, and more than a third worried about support after an AI purchase. It's a trust-and-accountability problem, not a technical one.


  5. What are the best AI tools for planning a trip in 2026? 

    General-purpose assistants such as ChatGPT and Gemini lead (Phocuswright found standalone AI platforms are used by 64% of AI-using travelers), followed by travel-specific tools like Mindtrip and Priceline's Penny, and the apps Expedia and Booking.com now run inside ChatGPT.


Comments


Commenting on this post isn't available anymore. Contact the site owner for more info.
bottom of page