I Trusted AI With My Vacation. It Sent Me to a Canyon That Doesn't Exist.

AI Can Put Together an Itinerary, but Not a Trip

On the third day of the National Day holiday, West Street, Quanzhou. Chen Jia and her friends were walking back from Xunpu Village. The ride-hailing page on her phone spun for ten minutes. No driver accepted. The bus stop was packed. Electric scooters crept through the crowd. They walked back to the hotel. For this stretch of road, at four in the afternoon, AI had told them, "You can make it. There is enough time."

This was not the first time AI had said "you can." It had said it many times. When it said it, they were still in the hotel. The car had not been called. The line had not been queued. The crowds on West Street had not yet grown out of the screen.

I. Paper itinerary

Chen Jia is 28 and works in Beijing. For the Mid-Autumn Festival and National Day, she went to Fujian: Xiamen, Quanzhou, and Pingtan. It was her first time letting AI plan almost the entire trip.

In the past, she scrolled Xiaohongshu, checked Dianping, and looked up distances on a map. This time she was busy with work. She threw her flight times, hotel location, and the places she wanted to go to Doubao and DeepSeek and had them arrange each day's route.

She gave detailed requirements. She did not want to walk 20,000 steps a day. She did not want overly commercialized places. She wanted no more than three or four main attractions per day. She also told AI it was a National Day trip and to take crowds into account.

The itinerary AI gave was complete. What time to get up, where to go first, how long the taxi between two attractions would take, which area to eat lunch in. She revised it several times, deleting places that were too far and scheduling popular attractions for the morning.

Chen Ting had also used AI. In the first half of the year, she went to Qingdao and Weihai. AI arranged the schedule and transportation. Nothing went seriously wrong. This Mid-Autumn Festival, she went to Changbai Mountain. She gave AI more authority, starting with choosing a hotel. She ran the plan through Doubao, DeepSeek, and the travel assistants on Tongcheng and Fliggy. She had been to Changbai Mountain twice and had stayed at Wanda both times. In early September, while planning for Mid-Autumn, she wanted something different. She asked AI to find a more recommended resort nearby.

AI quickly produced a plan. It looked reasonable and orderly.

II. On the ground

On the first day in Quanzhou, AI scheduled Xunpu Village for the afternoon and West Street for the evening.

At nearly 4 p.m., Chen Jia asked AI: Do you still recommend going to Xunpu Village at this time? AI said yes. It told her how long the trip would take and how long they could stay. It gave a compressed itinerary. She checked the time. From 4 to 6 p.m. was the golden window for zanhua photos. They went.

When they arrived, the crowd was beyond imagination. The line for zanhua photos was long. On a holiday, it was easy to get a taxi in, but hard to get one out. That night, back to the hotel, they could not get a car. The bus was too crowded to board. It took a long time to get back.

The hotel was not far from West Street, so they still went according to the itinerary. The crowd was bigger than imagined. Cars could not move. Electric scooters could barely move. They had already walked for a long time, so they got off and walked. The popular shops on West Street had long lines. Finding somewhere to eat took far longer than AI had estimated.

Chen Jia later summed up the experience: AI knows where every place is, but it does not know what those places are like during National Day. It is good at making an itinerary that holds up on paper.

Something similar happened in Xiamen. They had planned to go to four places but visited only two. AI could not judge the traffic and crowds on a holiday. It did not know they were already tired and did not want to rush to the next attraction.

Each time, it quickly gave a new plan. The new plan also looked reasonable. Later they found that the most effective method was to ask the hotel front desk and taxi drivers directly.

What Chen Ting encountered at Changbai Mountain was a different kind of failure. AI recommended a resort that sounded nice. After arriving, most of the shops in the resort area were not open. AI had recommended a place that existed but was deserted. It did not tell her what the local businesses were like outside peak season.

III. Hallucinations and danger

If "not taking crowds into account" is merely a nuisance, some AI itineraries can bring danger.

In September 2026, three climbers used Google Gemini to plan a climb of Mount Shasta in California's Cascade Range. The chatbot's advice was seriously wrong. They reached the summit and continued until late at night. Climbers are advised not to continue upward after noon and to turn back instead. They eventually descended a steep slope in pitch darkness. They called the Siskiyou County Sheriff's Office to ask for directions. Then they chose to deviate from the route and entered an adjacent canyon. They spent the night there. One of them injured a knee. The next morning, local search-and-rescue volunteers and U.S. Forest Service rangers escorted them back to safety.

The sheriff's office said Gemini had advised the three to carry far less food and water than the group needed. What was supposed to be an 8-hour climb turned into a grueling trek lasting several days.

Earlier, two tourists followed an AI recommendation to the "Sacred Humantay Canyon" in the Peruvian Andes. They spent $160 on a long taxi ride and discovered that the canyon did not exist. AI had combined two real place names and invented an attraction. A nearby guide overheard their conversation and advised them to stay away from danger. The guide told the BBC: "Altitude, climate change, and the passability of routes all have to be planned for. When you use a program like ChatGPT, it combines images and names to create fantasies. You may find yourself at an altitude of 4,000 meters, with no oxygen and no phone signal."

There have also been cases in China. A family of five in Hebei followed an AI-recommended "wild mountain route" into an undeveloped mountainous area. One person slipped and was injured. Firefighters had to enter the mountains for a rescue.

Some of these trips ended in wasted time. Others ended with rescue teams.

IV. Who is recommending

Even when safety is not involved, AI recommendations are not necessarily neutral.

Investigations have found hidden advertising in AI travel itineraries. Platforms use algorithms to prioritize "paid partners," "high-commission merchants," and "products from their own supply chains." This advertising is more covert than on traditional search platforms. Users assume AI recommendations are objective. Expert tests found that a hotel recommended by ChatGPT was 40% more expensive than another option with equivalent conditions three blocks away. These hotels had actively engaged in digital marketing targeting AI training keywords.

AI makes mistakes. Its tone remains sure.

V. The human steps

From these experiences, several steps that AI cannot replace in the short term become visible.

Pre-trip information verification. Opening hours, whether reservations are required, whether holiday crowd limits are in place, whether flights have been adjusted. This information changes constantly. AI struggles to grasp it in a short time. Travel agency staff say whether a hotel can accept a group, whether an attraction is normally open, and whether a flight is on time all depend on staff checking item by item.

Real-time traffic and road conditions. The drive times AI gives are often estimates under ideal conditions. Traffic jams, construction, and winding mountain roads are not in its calculations. One tourist returning from western Sichuan according to an AI plan was told 200-plus kilometers. The actual distance was 400 kilometers. During the National Day peak, doubled drive times are the norm. This part must rely on real-time maps and local experience.

Localized experiences. The small restaurants only locals know, the secret routes that avoid crowds. These require lived experience. Chen Jia found that the most effective method was to ask the hotel front desk and taxi drivers. AI is good at knowledge maps, not life maps. It can list attractions and food. When the question is grounded in the spatial structure of the real world, such as routes, directions, distances, and detours around traffic, it easily goes wrong.

Emotional interaction and handling emergencies. AI can handle standardized explanations. It cannot create a surprise on a tourist's birthday. It cannot adapt quickly in an emergency. Complicated road conditions, altitude sickness, itinerary changes. These require human flexibility and emotional reassurance.

Safety risk assessment. Outdoor hiking and mountaineering involve safety risks. Routes recommended by AI may contain hidden dangers. Safety decisions must be made by people with professional judgment.

VI. A collaboration framework

Whether AI makes itinerary planning easier or more chaotic depends on how it is used.

A practical division:

AI handles compiling attraction lists, building the itinerary framework, connecting routes, providing alternatives, and generating a first draft of the itinerary.

Humans handle verifying key information, judging real-time conditions, flexibly adjusting the itinerary, uncovering local experiences, and assessing safety risks.

For specific practice, one can learn from the interviewees' experience. After using AI to make a preliminary plan, verify key points through official channels and locals before departure. Once at the destination, decide in advance only the first destination each morning. Arrange the rest flexibly after arriving. Treat the hotel front desk and taxi drivers as real-time information sources.

Chen Jia said she will still use AI for future trips. Having it organize attractions, compare routes, and put together a few days of itinerary is much faster than checking by herself. The basic information has hardly ever been wrong. Opening hours and locations are correct. Treat AI's itinerary as a list of attractions. Decide only the first place each morning. Figure out the rest after arriving. That makes the trip more relaxed.

VII. The next morning

Chen Jia took screenshots of the AI-generated itinerary and saved them on her phone. The next morning, she opened it only to glance at the first destination. Then she locked the screen. She went downstairs, walked to the front desk, asked what time breakfast ended, and casually asked: If we go to this place today, what time should we leave?

The front desk said, "Before eight. Any later and you'll hit traffic."

She nodded and put the phone in her pocket. The itinerary was still in it. She did not open it again.