Can you use ChatGPT to plan travel? It’s hilarious and can actually work

California: Travel planning correlates with happiness - so say a million articles, researchers and well-published scientific papers. So why is it so danged overwhelming?

Maybe it’s because we think we’re looking for something unique when in reality we don’t want to venture that far off the beaten path. Or because the internet seems so full of ideas, yet always points us to the same over-Instagrammed cliché. (It’s a real paradox.) Or because our dreams are bigger than our budgets, given current headlines.

Tired of seeing the flickering cursor in a Google search bar - and wondering what it was that I was actually searching for - I decided to enlist generative artificial intelligence. After all, Chat GPT-4, the latest version of Open AI’s chatbot, promises to iterate creatively with users in order to solve complex problems. Here’s how it tackled a variety of travel planning situations - for two alter egos I created to test the technology - and how it might help you, too. The answer is, as with anything, a truly mixed bag.

Survey: Are you taking more blended work and leisure trips this year? Which airport in the world do you think is the worst? Take our MLIV Pulse survey. It’s short and anonymous.

The Success Story

The mission: A relaxing, family-friendly trip with two (very) young kids

Score: 9/10

Most glossy hotels would rather tell you how they cater to honeymooners than to toddlers, despite the fact that millennials - their once-coveted demographic - now globe-trot lavishly with their rugrats. I threw ChatGPT a curve ball by asking it not only to find me five-star Caribbean resorts with kids clubs but specifically ones that would accept my 4-year-old; to my frustration, most kids clubs start at age 5.

Its first suggestion was a perfect bull’s-eye: The Four Seasons Nevis actually lets kids as young as 3 participate in its Kids for All Seasons programming. (Many of those activities are complimentary, in a further surprise!) When I dug around the resort’s website to verify ChatGPT’s suggestion, I could see evidence of a pink-hued playground on the sand (that my daughter would love) and a beautiful arts and crafts station at the kids club. Sold.

It also suggested Eden Roc Cap Cana in the Dominican Republic, where I actually tried to plan our spring break trip this year. We scratched it because the flights from New York were astonishingly expensive; the resort itself is an amenity-packed dream for families with kids, and its Koko Kid’s Club indeed takes 4-year-olds.

They scratch the kid-friendly itch beautifully, but aren’t in the same threshold for luxury. On another query, ChatGPT recommended great hotels such as Malliouhana in Anguilla, where my kiddo isn’t actually old enough to partake in the 5-and-up Mini-Explorers’ Program. (Disclaimer: None of this information is easy to find online; I find myself digging for it in the least-seen corners of hotel websites. It’s also possible that ChatGPT was simply referring to outdated, pre-2021 information; that’s one thing that OpenAI warns about explicitly when you begin using it.)

The bot also did surprisingly well when I gave it even fewer parameters. In a fresh query, I asked it to brainstorm a relaxing vacation that I could take with my 1-year-old, ideally within two time zones of home. Since it didn’t know I live in New York, the second half of my question threw it for a bit of a loop”-prompting suggestions of Vancouver! San Diego! ChatGPT is human-like in the way it phrases responses, but it doesn’t ask follow-up questions. It’s best to be specific. Even so, its suggestions included Costa Rica and its rain forest reserve near the Arenal volcano, where I happen to have a future trip in the works.

ChatGPT didn’t quite drill down to the hotel I’m looking at (Nayara Tented Camp) but it came close by suggesting I go to the Four Seasons Resort Costa Rica at Peninsula Papagayo, where we could try “a range of family-friendly amenities and activities, including a kids’ club, family pool, and a variety of outdoor adventures such as zip-lining and surfing lessons.” (Did it remember I also have a 4-year-old and a taste for luxury?) The suggestion was a lucky one, I think: Papagayo happens to be two time zones away.

The Miserable Failure

The mission: A mental health escape on a $750-per-night budget

Score: 2/10

“I need a total mental reset,” I confessed to ChatGPT. “Can you find me a great yoga retreat in a luxurious resort, for May?” The query was rooted in wishful thinking: It’s a question for which I’d love to know the answer”-if time away from my family (and work) were even halfway possible.

Unfortunately, ChatGPT made the whole premise feel even more outlandish (sigh) when it recommended I head to the remote destination spa Ananda in the Himalayas or to a few posh spots in Bali - so very Eat, Pray, Love, I thought. I rephrased the question: Anything in Europe or the Caribbean? Sure, it said, and spouted off resorts in Sri Lanka and Thailand.

I started afresh. “I need a total mental reset,” I reiterated. “Where should I go on vacation?” This time, I got a slightly better-rounded set of suggestions, all destinations rather than resorts. (Costa Rica came up again; I should really just take that trip.) But I wanted hotel recommendations, so the bot came back with classics: Ranch Malibu, SHA Wellness Clinic, Kamalaya Koh Samui.

“My budget is $750 per night,” I followed up. “Do any of these work?”

What ensued was pure comedy. “Yes, there are some options that would work with your $750 per night budget,” it told me. “Here are some examples.” The Ranch Malibu typically costs $1,050 per night, it said, but a cheaper package might bring the price to $1,114. SHA works out to $826 per night. And at Como Shambhala, in Bali, terrace suites start at $815 per night. “This means that you could stay in a Terrace Suite for less than $750 per night,” it explained. I don’t think it was being sarcastic.

The In Between

The mission: A crowd-free trip to Europe

Score: 5/10

Here’s a trip everyone is asking me about: “How do I do Europe this August without the crowds?” I relayed the question to ChatGPT. Its answers were generally sensible: Scandinavia, Portugal and Turkey were all on the list. I wondered about the latter and pressed further.

“Isn’t Istanbul crowded in August?” I asked. “What’s the weather like?” It answered my questions with standard information that I might find in the front section of a guidebook. Ditto when I asked about cultural experiences: I got low-hanging fruit, such as going to a hammam or taking a cooking class, without specific recommendations.

But when I asked about destinations to visit in Turkey beyond Istanbul, ChatGPT got creative. Sure, it gave me a few obvious answers like Cappadocia and Bodrum, but it also suggested Antalya and Trabzon, a small but picturesque city on the Black Sea that I had never heard about and couldn’t find written up in any travel magazine.

The Verdict

Am I likely to go to Trabzon or to recommend it to others? I can’t say so. I just don’t know enough. And neither does ChatGPT.

While the engine recommended a few markets where I could theoretically buy handicrafts and artisanal foods, as well as a handful of the city’s higher-end hotels, it was never going to make me feel confident booking a trip to an unknown spot halfway around the world. Given its inability to do basic math, AI doesn’t command enough trust for me to embrace recommendations that entail such high stakes.

Which brings me to a point that I don’t see changing soon. When it comes to travel, it doesn’t make sense to trust anything automated or generic: That’s why we still find ourselves going back to the pros. What we’re looking for is happiness, right? This means something different to each of us.

Yet, as a preliminary planning tool, chatting with the bot was more satisfying than taking to Google and clicking on endless slideshows offering the best hotels in so-and-so places. When it works, ChatGPT’s randomness fosters a sense of discovery - which is what travel planning is all about. And when it doesn’t, well, at least it makes you laugh. The catch: You’ll have to do a lot more Googling to find out which one’s which.