As the National Day holiday entered its third day, long queues formed at scenic spot entrances, and high-speed rail stations and trendy districts were once again packed with crowds.

Many people took advantage of the break to travel, and before any trip, planning is unavoidable.

You open social media, save a few posts promising "a local's exhaustive guide," "a three-day two-night route with no detours," or "this one article is all you need," then drop each attraction, restaurant, and hotel onto a map to compare distances, prices, and reviews.

You still need to confirm which shops are open, what time the scenic area opens, and whether reservations are required during the holiday.

Often, before the itinerary is even finished, half the excitement for the trip is already gone.

In the past two years, more and more people have handed this task to AI.

At first, most used general-purpose large models like Doubao and DeepSeek: input your departure time, hotel location, and desired destinations, and it quickly arranges a route, specifying what time to leave each day and where to eat.

Later, travel platforms like Ctrip, Tongcheng, and Fliggy launched their own AI travel assistants, allowing users to check attractions, compare hotels, and book tickets all within a single chat window.

With more tools available and AI-generated itineraries looking increasingly polished, people naturally grew more confident.

But once the crowded long holiday arrived, the problems surfaced: AI can quickly organize public information, but it doesn't know holiday crowds or traffic conditions, can't accurately predict weather, and has no idea which shops are still closed.

It certainly won't adjust later plans based on your energy, mood, or unexpected situations.

We spoke with five people who used AI to plan trips during this Mid-Autumn and National Day holiday.

None of them were first-time AI itinerary users.

Their plans were detailed, but the pitfalls they hit were just as precise: a 50-minute drive turning into nearly two hours, a trendy district where almost no shops opened in the morning, and a pre-dawn summit wait for a sunrise that appeared for only 30 seconds.

Is handing your trip to AI a convenience or a hassle?

Their experiences may offer some reference.

A detailed AI itinerary that ignored National Day crowds

Chen Jia, 28, Beijing. This Mid-Autumn and National Day, I traveled to Fujian with two friends, mainly visiting Xiamen, Quanzhou, and Pingtan.

It was my first time letting AI plan almost the entire trip.

In the past, I would browse Xiaohongshu, check Dianping, and look up distances on maps one by one.

This time I was too busy with work, so I just threw my flight times, hotel locations, and desired destinations at Doubao and DeepSeek and let them arrange each day's route.

My requests were quite specific.

I didn't want to walk 20,000 steps a day, didn't want overly commercialized places, and wanted at most three or four major attractions per day.

I also specifically told it we were traveling during National Day and hoped it would factor in the crowds.

The itinerary it gave was extremely complete.

What time to wake up, where to go first, how long the taxi between two attractions would take, which area to eat lunch in 鈥?everything was laid out.

I had it revise several times, removing places that were too far and scheduling popular spots as early as possible.

On paper, the plan looked flawless.

It only fell apart once we arrived.

On our first day in Quanzhou during National Day, it scheduled us to visit Xunpu Village in the afternoon and stroll through West Street in the evening.

From a photography and distance perspective, that seemed reasonable.

It was already almost 4 p.m., so I asked AI whether it still recommended going to Xunpu Village at that time.

It said yes, told me roughly how long the journey would take, how much time we'd have to explore, and even listed a compressed itinerary.

I checked the timing myself, and 4 to 6 p.m. really is the golden hour for flower hairpin photos, so we went.

When we arrived, the crowds exceeded all imagination.

The flower hairpin photo queue was long, and while getting a taxi into the area during the holiday was easy, getting one out was extremely difficult.

That night we couldn't get a taxi back to the hotel, couldn't squeeze onto a bus, and it took forever to get back.

The hotel wasn't far from West Street, so we still went as planned.

The crowds were even worse than expected.

Forget cars 鈥?even electric scooters could barely move.

We were already exhausted and had no choice but to get off and walk.

The popular shops on West Street had long queues, and finding a place to eat took far longer than AI had estimated.

That's when I realized the biggest problem with AI itineraries: it knows where every place is, but it doesn't know what those places are actually like during National Day.

It's only good at making a plan that "works on paper."

Something similar happened in Xiamen during the Mid-Autumn holiday.

We originally planned to visit four places but only made it to two.

AI can't judge the traffic and crowds on a holiday, let alone know that we were already exhausted and had no desire to rush to the next spot.

It could always quickly give me a new plan, and the new plans all looked reasonable.

But we later found that the most effective approach was simply asking the hotel front desk and taxi drivers.

Still, I'll keep using AI for future trips.

Having it organize attractions, compare routes, and build a multi-day framework is genuinely much faster than doing it myself.

And the basic information was almost always accurate 鈥?opening hours and locations were correct.

Treating AI's itinerary as a list of attractions and only deciding each morning where to go first, figuring out the rest on the ground, actually makes the trip much more relaxing.

Letting AI help me "try something new," only to find most shops closed

Chen Ting, 36, Beijing. This wasn't my first time using AI for trip planning.

Earlier this year, I used it for trips to Qingdao and Weihai, and the scheduling and transportation planning were basically flawless, so I figured it could handle this.

So for this Mid-Autumn trip to Changbai Mountain, I gave it more trust and more authority, listening to it from the hotel selection onward.

I used Doubao, DeepSeek, and the travel assistants from Tongcheng and Fliggy to each make an itinerary.

I'd been to Changbai Mountain twice before and stayed at Wanda both times.

When planning the Mid-Autumn trip in early September, I wanted to try something different, so I asked AI to find a more recommended resort nearby.

It produced a comparison table, with its top pick being a resort that opened in 2021, 400 yuan cheaper per night than Wanda, with bigger rooms and newer facilities.

The most critical row said it was 38 kilometers from the North Slope scenic area, about a 50-minute drive.

My husband asked if the new place was reliable, and I said look at the reviews, and we booked it.

The first half of the trip went smoothly until after a night of rest, when we headed to the scenic area the next day and discovered AI's plan wasn't reliable.

The resort was huge, and getting from the exit back to the main road required a long detour on a side path.

AI said 50 minutes, but the driver actually took 1 hour and 50 minutes.

The straight-line distance did look short, but AI didn't have clear information about road conditions.

I'd encountered this problem before when using AI for trip planning and thought it would have been solved over the past six months, but it hadn't.

The resort itself was also different from expectations.

Many storefronts on the main street had "coming soon" barriers up, the supermarket wasn't open, and the entire resort had only two restaurants open in the evening, both serving cuisines we weren't interested in.

In the end, we ate expensive hotel meals for two days.

On the last day of the holiday, coming down from Tianchi, I asked AI what restaurants were open nearby, and it answered quickly and confidently: "Today is September 4th, early autumn..."

I stared at "September 4th" for several seconds.

That was the day I made the itinerary.

It seems to have no concept of time.

Every time I say "today," it easily gets confused about what day it actually is.

Looking back, the pitfalls this time weren't huge, but they wasted a lot of time, and I blame myself for being too trusting and not verifying.

Letting AI help you "try something new" is indeed risky.

Mature scenic areas and well-developed facilities cost a bit more but are worry-free 鈥?that's a calculation AI can't figure out.

For future trips, I'll still hand the itinerary framework and attraction comparisons to AI 鈥?it really is fast 鈥?but I'll definitely handle details like business hours, ticket release rules, and road conditions myself.

Also, many AI agent assistants now advertise long-term memory, claiming they can remember past mistakes and avoid them next time, or even book tickets and hotels for you.

But in practice, I find the barrier to use is still quite high.

I want traveling and trip planning to be relaxed and enjoyable.

If I have to spend time "training" the AI, the time cost is too high.

An itinerary that looked fine until I opened a map

Lingling, 25, Shanghai. I only started using AI for travel planning this year.

A friend recommended it, so I thought I'd try it and see if it could save some time.

My expectations were high at first 鈥?I thought with this, maybe I wouldn't have to think at all for future trips.

For this year's May Day trip, I had AI make an itinerary for the first time.

The content was detailed, with weather forecasts and travel tips, but the recommended attractions didn't match my preferences.

Some were too niche, and after reading about them, they didn't seem worth visiting, so I didn't use them.

Later, I used AI more as an encyclopedia during trips.

At attractions, I'd ask it to explain the history; when I wanted to know about local customs and specialty dishes, I'd ask directly.

At museums, it could introduce artifacts and historical figures 鈥?whatever I wanted to know, I could ask anytime.

Especially now that you can communicate by voice without constantly typing on your phone, it's very convenient.

Although I didn't use the first itinerary, I was still willing to keep trying AI.

This Mid-Autumn, I went to South Korea with a friend.

It was our first time there, and we weren't familiar with the area, so we had DeepSeek and Doubao help make a travel plan.

This time, I clearly felt it had improved. The itinerary was detailed and looked quite professional.

I let my guard down and didn't think to check it against a map before departure.

After arriving in South Korea, on the way to the hotel, I glanced at a map and realized the route had problems.

When it arranged routes, it seemed to have no concept of distance.

Attractions that were close together were split across two days, and we had to visit the same area twice.

For example, on the day we went to Dongdaemun, after browsing we could have visited nearby attractions, but it scheduled Lotte World far away, wasting a lot of time on the round trip.

Time allocation was also unreasonable.

Gwangjang Market, in my view, is basically a food market 鈥?people go for breakfast and a quick look around, not much time needed 鈥?but AI allotted a whole morning for it.

These problems aren't easy to spot when just reading the text.

Only by opening a map and cross-referencing the locations did the flaws become visible.

Route problems could still be caught in advance, but another pitfall wasn't so lucky.

The itinerary had us go to Hannam-dong one morning, but when we arrived, almost no shops were open 鈥?most don't open until noon.

We didn't even get to eat, and the trip was wasted.

After the Mid-Autumn pitfalls, I changed my approach for National Day.

I still wasn't comfortable handing everything to AI, so at the very least, I had to choose where I wanted to go myself.

I'd start with Xiaohongshu's "Diandian" to see what everyone was discussing under the same topic, pick out places that interested me, then tell AI those locations and have it string them into a route without backtracking.

If I wasn't sure, I could have it mark each day's stops on a map, then check it myself to confirm the route made sense.

This approach produced itineraries that suited me better.

Some friends of mine specifically look for travel planning Skills, and they've also made satisfying itineraries.

I think it's worth trying, but you need to spend time finding and filtering them first.

Finding the right method does take effort, but compared to problems during the trip, that time is worth spending.

Only by finding a way that suits you can you maximize AI's usefulness.

I trusted AI's sunrise time and got only wind on the summit

Xiao Feng, 28, Nanjing. This Mid-Autumn, I took an AI-generated schedule to climb Taizi Peak in Lin'an to see the sea of clouds.

It listed the sunrise time as confidently as a train schedule 鈥?05:49, not a second off.

What I got on the summit was basically just wind.

I'm not an AI developer, but I'm basically an "AI person."

Writing copy, making images, organizing travel guides 鈥?I throw everything at AI first.

My social media account, "AI Tinkering Office," has accumulated quite a few AI-generated results, and quite a few failure records.

The Taizi Peak trip added another entry to the case library.

My earliest use of AI in travel was a road trip to southern Anhui.

Searching for attractions, restrictions, and parking one by one was too troublesome, so I used AI for those chores and casually arranged the itinerary.

But back then, I only gave it simple tasks like looking up information, checking times, and ordering stops.

For Taizi Peak, I gave AI more authority.

The 288-kilometer night drive relied mainly on highway pilot assist, with me watching the road the whole time.

On the mountain, it laid out a complete schedule 鈥?what time to leave, what time to reach the trailhead, what time to start climbing, what time sunrise would be 鈥?and attached a gear list including headlamp, windbreaker, and thermos.

But it never mentioned the word "cloud" once, nor did it offer any alternatives.

It completely skipped the most critical premise: "whether you can see it."

I believed it, climbed to the summit in the dark, and the wind was so strong I could barely stand.

At 05:49, the sun peeked through a gap in the clouds for about 30 seconds and then disappeared.

The sea of clouds I wanted never came.

This may not sound like a major incident, but at the time I didn't feel that way.

After the cloud gap closed, there was nothing to see on the summit.

The wind blew in gusts, and once that moment passed, the rest of the schedule became useless.

AI had never planned for "what to do if you don't see the sea of clouds."

The real twist came on the way down.

I started descending around 8 a.m., and about half an hour later, passing through a thin layer of mist, a sea of clouds suddenly spread out before me 鈥?right at mid-mountain, right at my feet.

I realized then that the clouds I'd been waiting for on the summit hadn't gone anywhere.

They had just dropped to mid-mountain.

The sea of clouds I missed on the summit was fully compensated on the way down.

Honestly, I got lucky this time, enjoying a sea of clouds for free.

But this trip exposed the pitfalls of AI itineraries: it handles chores well, but never proactively deals with common-sense uncertainties.

Sunrise for AI is an astronomical calculation; for humans, it's a combined result of weather, cloud cover, and visibility.

Humans naturally know "you might not see the sea of clouds"; AI assumes "it happens on schedule."

By now, my division of labor is basically fixed.

Building frameworks, estimating driving times, and organizing scattered information into charts 鈥?I give those to AI. It's always available and never complains.

But what AI can't do is local experience and handling the unexpected.

It doesn't have local memory like "what time fog rises on this mountain" or "which section of road often has landslides," and it's completely useless for mountain trails, elevation gain, and physical stamina.

I've also developed three rules for how to use it.

First, don't take what AI gives you at face value, especially for tickets, reservations, and opening hours 鈥?always verify yourself.

Second, when asking AI, add a caveat and include the "what ifs," like "what if it's cloudy" or "is there a backup if tickets are limited."

Third, before departure, check the official WeChat account, scenic area phone number, maps, and ticketing platforms, and defer to official information.

I think "P people" (spontaneous types) are suited to using AI as an assistant 鈥?they're relaxed and tolerant, and don't get worked up when plans change.

If you're traveling with a "J person" (planner type), never rely entirely on AI. One mistake and it's a fight on the road.

So my feeling is: when using AI for travel planning, make the big decisions yourself, let AI handle research and chart-making, and double-check the key points yourself.

AI can't grasp the most real-time, localized information

Xiaolin, 32, Beijing. AI has entered many corners of my life.

What to eat every day, how to match clothes, even comparing prices before shopping 鈥?I'll casually ask it.

So I once thought AI should be especially suited for travel planning.

I'm a typical "P person."

Before, when traveling, I'd also browse many local guides in advance and save restaurants, attractions, and malls to my bookmarks, but it was hard to actually produce a detailed itinerary.

Usually, I'd decide where to go on the day, and if I wasn't in the mood, I'd lie in the hotel all day.

My initial expectation for AI was that it could organize scattered information into a truly executable itinerary, plan routes based on attraction distances, and compare flight prices to save some money.

In June this year, I went to Thailand with a friend and for the first time seriously involved AI in the entire trip preparation.

Before departure, I had Codex search flight prices daily to find a better time to buy.

After a few days, I found that aside from consuming tokens, it had almost no advantage.

It wasn't faster at finding information than me checking Ctrip or Qunar myself. Giving it to AI actually added a layer of communication cost.

Later, I had Claude make a detailed road book.

At first glance it looked very professional.

How to go through immigration at the airport, how to get a taxi, how to get a SIM card, what time to leave each day, where to go in the morning, where to go in the afternoon 鈥?it even included a few common Thai phrases.

But once in Thailand, I quickly found the itinerary was almost impossible to follow.

The most obvious problem was that it had no concept of "travel experience."

It crammed too many attractions into each day.

Some places looked close on a map but were actually troublesome to get to.

Neither my friend nor I like "special forces-style" travel, so if we lingered a bit in the morning, the entire afternoon's plans were scrapped.

In the end, we had to renegotiate where to go the next day almost every evening.

Some of the most practical-looking information also didn't hold up in practice.

One day we planned to visit the Grand Palace, and ChatGPT earnestly recommended a ticketing platform, saying it was convenient and cheaper.

I downloaded a new app, registered, searched for a while, and found the tickets on that platform were sold out.

My friend opened Ctrip next to me and not only were there plenty of tickets left, the price was even cheaper.

That moment made me suddenly understand the most awkward thing about AI itineraries.

It has extremely strong information organization skills, but it doesn't necessarily have the most real-time, localized information, and it struggles to understand a person's true preferences during travel.

By this Mid-Autumn, when I went to South Korea, I had learned to use AI in a limited way.

I still let it recommend attractions, restaurants, and areas to visit, but I no longer let it complete the entire trip for me.

From the dozens of suggestions it gives, I pick the few that genuinely interest me, then check reviews, distances, and business hours on Xiaohongshu, maps, and booking platforms.

AI is great for expanding the boundaries of information, but not for deciding where I should go.

Travel is ultimately a very personal matter.

AI can tell you what exists in the world, but what's worth seeing for yourself still requires your own judgment.

*The cover image and illustrations in this article are from Pexels. At the request of the interviewees, Chen Jia, Chen Ting, Lingling, Xiao Feng, and Xiaolin are pseudonyms.