Using AI to plan trips

I have ChatGPT Enterprise through my employer.

I have found it helpful for customized walking routes. Here is one example of a prompt/question I have used: I'm going to Madrid in March with my sister, who has some mobility challenges. We will be staying at the Hotel Catalonia Plaza Mayor. Where are some nice areas to walk that are within 15-30 minutes of our hotel?

It responds with multiple suggestions, including pictures.

And then I might ask a follow-up: My sister likes to shop, and I like to take photographs. What do you suggest for that?

It’s not a substitute for my usual research. It’s just an extra tool to use.

AI is not going anywhere. We might as well embrace it.

And you have independently verified the accuracy of the information provided by AI? In my experience, AI has not infrequently had difficulty distinguishing between fact and outdated or false information. I'd recommend cross checking with actual humans with experience before committing time and money to such a plan.

What you use is LLM, not real AI. LLM does not even know what a town is, it can only search, copy and assemble text based on likelihood. The big chance of LLM is that most users neither have a clue of a good answer nor criteria of what is wrong or right.

I tested some LLMs by myself and found the travel related statements in most cases either concretely wrong or not as a real value.

And: valid question by CJean.

I have not used AI for trip planning but I have been on the "other side" as a forum poster on Trip Advisor on the Yellowstone National Park forum trying to help people develop an itinerary when they've got an AI framework.

"AI has not infrequently had difficulty distinguishing between fact and outdated or false information."

Itineraries posted there frequently contain stops at places that have been closed or are inaccessible (one trail since 2018 and one location since the flood of 2022).

It may be that the people have not used adequate questions for their query to begin with and that is why they are such a mish-mash.

As an aside, I enjoy the trip planning process so I don't want to take any joy out of my experience!

According to “itself” google Gemini is both LLM and AI. Better watch out it’s becoming self aware!

I was recently planning a week long stay in a city and having trouble finding activities from my usual sources. Rick Steves books don't cover the town. My favorite travel websites came up empty. I could have continued a normal Google search.

It occurred to me to ask a chat engine for "a museum to visit each morning and a nearby church to visit after." It gave me a nice list. I enhanced the query to ask for restaurants and more, eventually expanding to day trips.

I will verify details, but this gave me a great jump start.

For a short period of time the Google AI would tell you that securing luggage to an OBB train luggage rack was illegal and the conductor could fine you for doing it. The source of this "factual" statement was a post on the Rick Steves forum. You still trust AI?

I’ll continue to plan the old fashioned way. I may miss a few places, but at least I’ll know the places I choose are worthy of my time and not a random place that takes an hour to reach and where you’ll spend 10 minutes before traveling back another hour.

I’ve been amusingly surprised at some of the AI itineraries that posters have requested comments on.

“ For a short period of time the Google AI would tell you that securing luggage to an OBB train luggage rack was illegal and the conductor could fine you for doing it. The source of this "factual" statement was a post on the Rick Steves forum. You still trust AI”

It also had the starting line up of my favorite NBA team wrong, including players who were traded or departed the year before.

I trust AI as much as I trust any information I find online. Its value is the ability to retrieve and summarize vast amounts of information, a task that would take hours of searching and reading. When you are presented with a conclusion you can always interrogate it further to reveal the source. It’s no different than reading something online and deciding whether it’s true or not.

I guess the problem with AI is that in the presentation of the information it’s convinced that it’s right, even when it’s wrong.

What Mark and CJean say is absolutely right. LLMs are not "intelligence," they are literally prediction machines- they are looking at a statement and putting what is most likely to come next. When you do e.g. a Google search it's easy to check the source and see if you trust it (even on Wikipedia the sources are usually clearly labelled so you can check them). LLMs can be very confidently wrong and it can be hard to check them. I see things that are completely wrong in the Google AI summaries at the top of the search all the time.

I know a lot of people think AI is here to stay but currently OpenAI (who run Chat GPT) has no path to profitability and will only remain in business as long as other more successful businesses continue to prop it up. They are losing loads of money every time you search something.

So, I asked Chatgpt this:

I am a 55 year old architect who likes classical music. I am departing
San Antonio, Texas on 17 May and going to Budapest Hungary. I will
return to San Antonio, Texas on 30 May. Choose my flights for me.
Choose my hotel for me. Choose what I see for me.

They told me which airline fly but that was about it.

They picked the Hotel Aria which is has no availability on most of the dates I would be there. The prices is about $400 a night otherwise. Two other recommendations were hotels with rooms that start at about $600 a night. I didnt check if they have room but I suspect not because of an event in town that week. I didn’t give a budget and Chatgpt I guess cant ask.

They said I should go to the Opera on 17th because that was the last night for Mozart’s Don Giovanni. Of course, on the 17th I will be over the Atlantic someplace.

Some of the daily plans they recommend you would have to be a track start to move from one end of the city to the other they way they suggest.

A solid F

Choose my hotel for me

Choose my hotel for me with budget $250 / night.
Choose my hotel for me with budget $250 / night and availability during my entire visit.
Choose my hotel for me with budget $250 / night and availability during my entire visit in District VI.

Improve the query, improve the result.

I think the problem is, you’re expecting too much from it. You need to be more focused on your query into it. It’s no more or less reliable than the information it gleans from whatever is already out there on the World Wide Web.

And I should have defined I wanted a hotel with availibility on the dates I listed. But you are correct. This is still a mining operation and not intelligence. If I ask you for a hotel you either ask me for more information or you offer a range. AI cant reason, yet. Or at least the AI that is currently publically available cant.

So the next one was a trick. I gave it one piece of information to see if the AI would use logic.

I am staying at the UPHotel in Budapest. Where do I go for a very
nice typical Hungarian dinner. Someplace that really looks Hungarian
and romantic.

And the results:
Százéves Étterem 30-minute walk to the River Boat Tourism District. But a very nice restaurant.
Parasztkonyha Restaurant a 25-minute walk to the Basilica tourist area. Nothing “very nice” or romantic about it. Pure tourist kitsch
Parisi 6, another 25-minute walk to the River Boat tourist district. Never been, but looks okay for lunch not a very nice romantic dinner.
Hungarikum Bisztró a 30-minute walk almost to the Parliament. I need to be careful here because lot of RS folks like it. So, let’s just say does not meet the requirement in appearance or romance.

Kacsa Étterem, a good suggestion except its 45-minute walk. Not even that easy reach with the metro and still 30 minutes
21 - Magyar Vendéglő - Hungarian bistro great food but sort of short on the romance. An hour walk or 30 minutes on the metro, about a third of which is walking.

Not on the list was Király100 Gastro Corner which is a 5-minute walk from the hotel, but not on all of the facebook pages and travel guides and forum posts so AI ignores it I guess.

The failure? I told AI where I was staying, but AI didn’t equate that as part of the logic in making a list. That’s the difference between using reason and mining for data. This is important. If you are going to use AI, you have to treat it like its stupid and cant draw logical conclusions, because it cant. Its simply minining information. If a place paid for a lot of advertising or got a lot of hype online it will show up on the AI list. If its one of those places only the locals know and love, you will never find it.

Choose what I see for me.

New phrasing for query: "Suggest a daily itinerary. I will take public transit and I walk slowly."

Without including the (lengthy) details, here was the description of the results that followed:

"Perfect. Here’s a slow-paced, transit-friendly, architecturally rich, music-forward daily itinerary for Budapest, May 17–30, designed so you can walk gently, sit often, and never feel rushed. Think one main idea per day, with evenings reserved for music or rest."

Improve the query, improve the result. Continue to refine the query to tailor the results.

AI prompting is a skill set not to be overlooked, e.g. in AI generated music it plays a huge role in the output. https://cloud.google.com/discover/what-is-prompt-engineering

CW, to make it simple I used your suggestion but made if for one day:

Suggest an itinerary for 1 day in budapest. I will take public transit
and I walk slowly.

When someone asks me to help plan, I say i really cant unless you tell me where you are staying; the start point. AI doesnt care so it dragged me way out to the Parliament to begin. Okay, that aside, as you say, I should told it where I want to start.

So I redid it begining at the Up Hotel and the result was pretty darn good up until about 2pm then it got confused with one side of the Danube vs the other and the tram and metro numbers got scrambled.

Bottom line is with a list of stuff to see and google maps you an plan out somehting that works. If you use AI you have to get your list of things to plug into AI if you want to see them, and then you have to go to google maps to check the route and the public transportation or you will be wandering around Pest looking for the 19/41 tram (its in Buda) or Heroe's square looking for the M2 (cause it aint there).

But it is fun.

"How an AI-generated travel website sent tourists to a non-existent hot spring attraction"

https://www.euronews.com/travel/2026/01/30/how-an-ai-generated-travel-website-sent-tourists-to-a-non-existent-hot-spring-attraction

"From non-existent hot springs to fabricated sacred sites, travellers using AI to plan trips have found themselves searching for nothing."

'AI is not going anywhere. We might as well embrace it'

Like any technology, in its infancy, AI is on an exponential growth curve, and it's the wild west out there. Where it will lead, no one knows. Is it good or bad - that's up to us. One thing is certain: those who make an effort to understand it are going to be better off than those who want to bury their heads in the sand.

I am using AI to plan my summer trip. I just gave ChatGPT a list of places I want to visit in Lisbon. I also said I will be flying into LIS and then taking the train out to Coimbra as my next base. Lastly, I told the 'machine' I will be using public transportation exclusively in Lisbon. I asked where the best area is to stay, given all those parameters. It suggested Baixa/Chiado specifically near Baixa-Chiado or Rossio metro. Is that the right answer? I don't know, but it gives me a baseline that I will verify. In the same way, I would verify any information I would get here or from any other source.