What happens if you use AI to plan your next vacation? The answer is not always good.
Artificial intelligence is not only taking over the internet, it’s making inroads in the travel planning industry. But is the information you get from free and paid AI tools like ChatGPT-4 and Bard and apps like Vacay, Trip Planner AI, Roam Around, and iPlan.ai reliable? Definitely no, says Heather Stimmler, an expat living in Paris who wrote Fodor’s France guides from 2005 to 2015 and has found that AI-generated tour guides of the classic city are way off the mark.
“I’ve made hundreds of customized Paris itineraries over the past two decades and even simple ones are harder and more time-consuming to make up than you’d think,” she says. “A good itinerary not only takes into account what a visitor wants to see, but also if it’s feasible, how the person will get there, when a destination is open and if reservations are needed, among other variables. And AI tools fail at most of these tasks.”
How Do AI Planning Tools Work?
AI planning tools cull their itineraries from a mix of licensed and publicly available information and data input by human trainers. ChatGPT-4, which was launched in November 2022, says it generates responses based on perceived patterns and the information available to it, although its database is only current as of 2021 (which, of course, was during the pandemic, when the world was in tumult and many places were closed or have since gone out of business).
Top Picks for You
Recommended Fodor’s Video
In reference to ChatGPT-4, Stimmler says, “The problem is that it answers questions it doesn’t know the answer to instead of saying it doesn’t know the answer!”
What’s more, it is notorious for producing nonsensical answers. And by the way, it’s not just itineraries AI gets wrong—AI-generated guidebooks are no better. The New York Times recently reported that shoddy and inaccurate guidebooks are also flooding the market.
Testing the Bots
For her website Secrets of Paris, Stimmler recently tested ChatGPT-4, Bard, and several apps that generate AI itineraries. “The results were worse than I expected, with many mistakes,” she says. Besides spelling and address errors, the tools would send people to places like Notre Dame Cathedral, which can’t be visited right now and is under scaffolding since a devastating fire almost took it down in 2019.
She found that the directions were often incorrect and in fact, “that’s when things got ugly. If you don’t know a place, you’ll waste your time and be frustrated when AI sends you the wrong way or doesn’t account for how long it takes to get from one place to another.”
While she suggests that the problem may be more of an issue with large cities like Paris where things are constantly in flux rather than smaller destinations like, say, Fort Wayne, Indiana, she recommends using Google or Google Maps for planning instead. At the very least, if you use an AI planning tool, be sure to check everything it says against more reliable sources, she urges. That’s what Sam Gaspar, of Redding, Connecticut, did when he was planning a last-minute trip to Puerto Rico recently.
“I normally like to research and plan trips out in detail myself using an Excel spreadsheet, but I didn’t have time to do that this trip so I used ChatGPT-4,” he says. “I asked it to give me a week’s itinerary. I told it to assume I had a car and I didn’t want to travel more than two hours a day. After that, I asked more specific questions.”
Gaspar ended up using the AI-generated results as a blueprint for his trip but didn’t take the results at face value. “I would use ChatGPT-4 again to plan a trip but I wouldn’t use it as my only source,” he adds. “I recommend matching the AI information to user reports on other websites to make sure it’s accurate.”
I, too, love planning trips, so I was intrigued at the idea of using ChatGPT-4 myself (although AI also terrifies me with visions of the Arnold Schwarzenegger Terminator movies and the end of human existence dancing in my head). I’m not familiar with the Midwest, so I tested it out to devise a route from Oklahoma City, Oklahoma to Mount Rushmore. It was easy and free to use the basic service, but I found that my initial question was overly broad and vague. Apparently, the AI is not the only one who needs to do some learning.
I started with: “Help me plan a trip from Oklahoma City to Mount Rushmore,” and I got a meandering five-day itinerary when I really wanted something faster. I then asked: Help me plan a two-day road trip from Oklahoma City to Mount Rushmore,” which the chatbot and I both agreed entailed too much daily driving. Finally, I asked: “Help me plan a three-day road trip from Oklahoma City to Mount Rushmore,” which the bot said would allow me to enjoy the journey and explore along the way. Unfortunately, it had me going the wrong way on I-80 (west when I should have been instructed to go east) toward Omaha, NE—not to mention that I didn’t want to make a stop in Omaha, which was out of the way on my proposed journey.
Ultimately, while it was fun to use AI, like Stimmler and Gaspar, I found that it wasn’t trustworthy. I ended up checking the route on a good old-fashioned printed Rand McNally Road Atlas and on MapQuest, which is how I found the ChatGPT-4 error. I also found that MapQuest has a superior route planner that can be customized for the shortest route for distance or timing and gives different options for routes accompanied by highlighted maps.
“You can also just use Google or Google Maps to research a trip, which I find to be more reliable than AI tools,” says Stimmler.