We have been testing ChatGPT, Perplexity, Google Gemini, Claude, and Meta Glasses as a digital travel companion over the last several months. For a recent spring break trip to London, we asked it to plan routes, recommend restaurants, and help structure a daily itinerary.

ChatGPT was the best performer overall, and Perplexity was a crucial backup, particularly for the most current information.

But like a well-meaning yet clueless friend, both had moments of brilliance – and then utter failure. For one task, it would surprise and delight us. The next task, it fell flat on its face. 

Let’s take a quick look at where AI gets it right, where it struggles, and how to use it without losing your mind.

The good: Better than brochures

When it comes to big-picture planning, AI is an insightful and reliable travel buddy. Ask it to give you the most important highlights about a landmark or place you plan to visit, and it'll provide a fantastic summary. 

Dig a bit deeper and ask ChatGPT for the most interesting facts and historical moments, which will unlock a treasure chest of trivia.

It’s also great at providing insights into neighborhoods – famous people who lived there – and uncovering pop culture-related locations, like the Blue Door address where the main character lived in “Notting Hill” – the movie. 

Prompt tip: Ask ChatGPT, "What would surprise me about (name the landmark or place you will visit)?”

The adequate: It tries, bless its heart

Recommending food and beverage locations? Mixed bag. ChatGPT tries, but it's not TripAdvisor, and it often misses the places that are closer in proximity. We asked many times to recommend a place – then tested it against standard Google searches, and Google crushed it.

While ChatGPT would provide an embedded map with the info, we just wanted a list with the exact details to scan quickly. This is a great illustration of where AI still massively fails – giving us more than we need or ask for and providing it in the format it wants, not how we want it.

The bad: Proximity and timeliness are not strengths

Try to use ChatGPT to retrieve time-sensitive info like hours, recent reviews, or current menus, and it stumbles. It also likes Yelp (which we hate) and can’t tell if a top review came from a real human or a bot.

Want a list of lunch spots near a particular museum or a dinner place within a 10-minute walk of your hotel? Expect hilarity. It’ll recommend restaurants and bars 30 minutes away – by car – emphatically suggesting that it is "nearby." 

Don’t ask it to map your day, or you’ll find yourself zigzagging like a tourist on espresso.

Real moment: We prompted every chatbot by sharing our list of places we planned to visit each day and asked it to map out the most efficient walking route. Every one of them came up with an itinerary that made absolutely no sense when you plotted them on a map.

The ugly: Terrible at travel times

The chatbots and Google Maps must share the same dataset to estimate travel times. When it comes to walking, they all must think I'm an Olympic athlete. Seriously, even then, travel time estimates – mass transit, car, you name it – were 100% wonky. Walk time estimates were by far the most error-prone.

We discovered that doubling walk times, adding 20-30 minutes onto mass transit times, and adding 10-15 minutes on Uber/Taxi times would repair the AI mistakes.

Use AI for travel knowledge, not as a guide

AI’s strength is insight. It’s a wickedly fast researcher, a good brainstormer, and a decent idea generator. But it fails as a mapper, sucks at thinking like a GPS, and definitely should not be your travel agent.

Use ChatGPT to start your plan, but use Google Maps and your favorite review sites to lock in your best routes – and choices. (-Kevin)