The Confident Wrong Answer

An older couple in Malaysia drove more than three hundred kilometers last summer to ride a cable car.

They had seen it in what appeared to be a television news segment, the sort of thing you would never think to question: a reporter with a microphone, visitors queuing at a ticket booth, mountain views, and a deer petting zoo waiting at the top. The Kuak Skyride, in a quiet town called Kuak Hulu. It looked like a local broadcast because it had been built to look like one.

When they arrived, they asked a hotel employee for directions and were told there was no cable car. There never had been one. The entire segment had been generated by artificial intelligence, down to and including the reporter.

The woman said she intended to sue the journalist, and the hotel employee had to explain that the journalist did not exist either.

What has stayed with me is the question she asked next, which was reported afterward. Why would anyone want to lie? There was even a reporter in the video.

This is not a story about gullible people. I want to say that at the outset, because the instinct is to read it as one, and I think the instinct is wrong. That couple did nothing careless. They watched what looked like a news report on a television channel and believed it, which is what most of us have been taught to do our whole lives.

Squaremouth, the travel insurance comparison site, surveyed more than 2,500 travelers in the autumn of 2025 about how they plan. Forty-seven percent had used AI to build an itinerary. Of those, thirty-four percent reported getting back information that turned out to be false or misleading. And forty-four percent said they trust AI as much as they trust a search engine, a friend, or a travel advisor.

Sixty-five percent said they would use it again regardless, and I think that is the most honest number in the whole survey. These tools are genuinely useful, and people can feel that they are useful. The trust moved because the usefulness is real. It simply moved faster than the accuracy did.

Where AI is actually rather good. I will say this plainly, because Schelli and I both use these tools and it would be dishonest of me to pretend otherwise.

AI is wonderful at the beginning of a trip, when everything is still vague and you are trying to work out what you even want. Ask what the Dolomites are like in early July, or whether ten days is enough for Portugal, or what really separates a river ship from a small ocean ship, and you will get a sensible, well-organized answer in seconds. It is a marvelous translator. It will build you a packing list, explain a visa category, or walk you through the shape of a region you have never seriously considered before.

It is very good indeed at helping you decide what kind of trip you would like to take. It is not good at telling you which trip is actually available, on your dates, at a price that is real.

Where it falls down, and why. It does not know what has closed. Judy Gauthier of Go City described being routed by an AI planner to a trail in the Great Smoky Mountains that had been shut for eighteen months, and only discovering it from a Facebook post. The model was not lying to her. It simply had no way of knowing, and nothing in the answer signaled that there was a gap. Restaurants that closed last spring, hotels in the middle of renovation, a museum that has been dark on Tuesdays since the new schedule came in. These are the most common failures and the least dramatic ones. You lose an afternoon, and you feel a bit foolish.

It invents things, and it invents them with great confidence. The BBC reported on two travelers in Peru who set out for a place called the Sacred Canyon of Humantay, having paid roughly $160 to reach a rural road near Mollepata. A local guide, Miguel Angel Gongora Meza of Evolution Treks Peru, overheard them and stopped them. As he put it to the BBC, “There is no Sacred Canyon of Humantay.” The name was a combination of two real places that had nothing to do with the description they had been given. They had been intending to hike into the Andes without a guide, at altitude, looking for somewhere that existed only in a paragraph. The same reporting described a couple stranded at the top of a Japanese mountain after being told the wrong hours for a path, and visitors who traveled to London for a Christmas market at Buckingham Palace that had never been planned at all.

The pattern in every one of those is the same, and it is the thing I would most like you to take away. The answer contained no hesitation. A person who does not know something will tell you they are not sure. A model that does not know will often say nothing whatsoever about not knowing.

It cannot see what sits behind the booking. This is the one I notice most, because it is the substance of my working life. Schelli and I have just spent a week aboard Star Princess, and one of the more useful things we came away with is that the staterooms directly beneath The Dome are lovely and quiet through the day, and the music runs late into the night. That is not published anywhere. It is not a review score or a label on a deck plan. You know it because you slept there, or because somebody who did told you. Multiply that by every ship, every property, and every connection that is technically legal and practically impossible, and you have a fair description of what this job actually consists of.

It has no standing when something goes wrong. This is the part almost nobody thinks about until it is too late to matter. Most travel insurance will not cover a disruption that arose from acting on incorrect information, and that includes incorrect information from an AI tool. If a model invents a departure time and you miss the flight, that is generally going to be your problem to solve and your money to lose.

And when the failure is larger than a missed flight, the gap becomes rather a chasm. CNBC reported earlier this year that when airspace closed across the Middle East and travelers found themselves stranded, no chatbot moved a single person up the queue for a repatriation flight. It could not telephone the airline. It could not telephone the hotel and hold the room for another night. It had nobody at all to call on anyone’s behalf, because it does not know anybody.

What a ship’s bridge taught me about all of this. On our last sea day in Alaska we were given a tour of the bridge of Star Princess, and I came away with one fact that I have not stopped repeating to people since.

I had assumed, as I think most of us do, that a vessel that modern more or less drives itself between ports. Satellite positioning, integrated navigation, redundant systems, screens in every direction. It does not. There is an officer at the controls and a full watch on duty every minute of every voyage, around the clock, in open water and in port approaches alike. The technology aboard is extraordinary and it is doing an enormous amount of work. It informs the decisions. The people are the ones who make them.

That is the whole of it, really. Not technology instead of people, and not people refusing the technology, but very good information placed in front of someone who is accountable for the call, who has been through that channel before, and who is still standing there at three in the morning when the weather turns.

I would not want to sail on a ship run any other way, and I do not think you would want your trip planned any other way either.

Four questions worth asking. If you are using AI to plan, and please do feel free to, these four will catch most of the trouble before it reaches you.

When was this last true? Ask the tool directly for its source and the date attached to it. If it cannot produce one, treat what it has told you as a promising idea rather than a settled fact.

What would you have told me if you did not know? Models very rarely volunteer their own gaps. Ask outright what it is uncertain about, and watch what comes to the surface.

Has anyone actually been here? Not whether it is well reviewed, but whether a specific person has stood in that room, made that connection, or sailed that ship in that particular season.

Who is accountable if this turns out to be wrong? This is the real question, and it is worth sitting with. If the answer is nobody, then you are carrying the risk yourself, and you ought to know that before you are standing at a locked gate with your luggage at your feet.

What I actually think. I am not here to tell you that AI is bad, because it is not. It is a remarkable first draft, and it has made the early, foggy part of planning a good deal easier for a great many people.

But a first draft is not a plan. Somewhere between the draft and the departure, someone has to confirm what is real, hold the space before it disappears, read the terms that decide what happens if you need to cancel, and be reachable on a Sunday evening when a flight falls apart.

That is the part we do. Not because we can out-search a machine, because we certainly cannot. Because we can be telephoned, and we can be held to what we have told you.

If you would like a real person to look over something you have been building, do bring it to us. Bring the itinerary, the screenshots, the twenty open tabs. We will tell you honestly which parts are good, and there usually are good parts, and which ones will not survive contact with the actual world.

Warmly, Kari · Kingfisher Luxury Travel

Sources: Squaremouth, “More Travelers Are Using AI Without Realizing the Risks,” press release, 21 October 2025. Survey of more than 2,500 travelers.

BBC Travel, “The perils of letting AI plan your next trip,” September 2025. Reporting on the Sacred Canyon of Humantay, the Japanese mountain path, and related incidents; quotation from Miguel Angel Gongora Meza, founder and director of Evolution Treks Peru.

Outlook Traveller, “The missing layer in travel AI: accountability when it matters most,” April 2026. Account from Judy Gauthier of Go City regarding the closed Great Smoky Mountains trail.

CNBC, “Travelers are turning to AI to plan trips — but hallucinations and trust gaps remain,” 11 March 2026.

The Kuak Skyride incident was first reported by Malaysian outlets and subsequently covered by the South China Morning Post and The Independent, July 2026.