Can AI Give You a Good Tour of Tulsa? We Put ChatGPT to the Test

I love a good mashup. Steampunk scenes with brass gears spinning inside steam-powered computers. A New Orleans brass band doing an AC/DC cover. 1960s cars with modern drivetrains.

So it wasn’t a surprise when the idea came to mash up the AI chatbot tools of today with our old full-time RV traveling selves.

Two things consumed most of our time back then. One was deciding where to go. The other was figuring out what to do once we got there.

Could AI make that easier?

I set up a scenario: Pretend we were in Tulsa on business. The week had been too busy for sightseeing. It’s now Saturday, and we have four to five hours to get a taste of what makes Tulsa, well, Tulsa before flying out tomorrow. It’s hot, we have a rental, so make it a driving tour. Figure $250 max for sightseeing, souvenirs, and lunch.

Could AI research locations, generate a downloadable driving route using Google Maps, and provide us with insightful audio narration?

I chose the most popular AI tool – ChatGPT. I didn’t give it any preferences on what to see. The point was to find out what places in Tulsa the AI “thought” was worth showing a visitor.

We named our guide Ferguson, after Mark Twain’s habit in The Innocents Abroad of calling every local tour guide by the same name rather than learning their actual name.

Round One: A Good Start

Ferguson was excited about the task. He quickly suggested that his role be bigger than just a simple Tour Guide. He wanted to be a Producer. Then Art Director, suggesting what photos to take. Then Graphic Designer, laying out the final article visually.

I demoted him back to Tour Guide, assuring him I had people for all of that. He took it well.

Ferguson suggested a flexible as-we-go itinerary. Find one stop, enjoy it, then find the next. Sounded reasonable. I pulled out our old Garmin GPS to handle navigation while letting Ferguson run the tour on the phone. I installed the ChatGPT app, then set up a voice agent to read aloud to us. For some reason an Australian accent felt right.

Ferguson’s first stop was a Boston Avenue address in the Art Deco district, right by the Atlas Life building.

All good – Tulsa is well-known for its Art Deco architecture. Ferguson narrated a brief history of the buildings and answered a few questions (like “Q: Why did they build tall buildings when there was so much open land.” “A: Ego.”)

But then he said we should be able to see the Mayo Hotel from there. Three blocks away. Behind a very tall building.

Strike one.

His next stop was the Greenwood Rising building in the Black Wall Street area. Again, an understandable choice. Anyone new to Tulsa should know about the Tulsa Race Massacre of 1921 and what’s happened in this area since then. We listened to his narration, took some photos, and asked for the next stop.

Route 66 was next. I expected Ferguson to take us to the “Land of the Giants” stretch of Route 66 on 11th Street, or maybe the Cyrus Avery bridge and sculpture.

Nope. He directed us to the 2500 block of East Admiral Place, near the intersection of Birmingham Avenue – a mixed industrial and residential stretch with little to see. When questioned, he said something about this section of the road being a “story of resilience” and admitted he “had likely valued a historically accurate road alignment over what a visitor would actually want to look at.”

Strike two.

From there it was to the Philbrook Museum – also an understandable stop on a good Tulsa tour. Ferguson suggested we just could park and “enjoy the gardens” – failing to mention the $40 entrance cost for doing so. We paid, enjoyed the gardens and indoor sights, grabbed a snack, and resumed our initial tour attempt.

Ferguson’s next suggestion was the Tulsa Air and Space Museum. Way back on the other side of town, closer to where we’d just been. I realized I had not instructed Ferguson to make sure his route was efficiently laid out. But, we gamely stuck to the experiment and drove over. This time we elected not to go in, took some photos in the parking lot, and asked Ferguson for the next stop.

He said since we were sitting right next to Guthrie Green, we should visit that next. I questioned his choice and he admitted we were miles away from the Arts District. I prompted for a different stop, and he directed us to the Tulsa Historical Society, but wrongly said it was on Gilcrease Museum Road, and then also failed to mention the road was closed for construction.

Strike three.

Still trying to finish strong, I asked for another stop. His next suggestions were places we’d just been. I gave up and asked for lunch suggestions. Not happy with his initial choices, I asked for alternatives – but he kept giving us the same three places over and over while describing them differently each time.

Afterwards, I noted there were two large omissions in Ferguson’s itinerary: Native American history – central to how Tulsa’s formation – never got a mention. And Tulsa’s deep music history simply disappeared with no explanation after an early mention.

And then?

Ferguson lost the entire transcript of the tour.

We gave up, and decided to try again the following day.

Round Two: A Tighter Plan

Round two started with a different approach. Instead of letting Ferguson improvise stop by stop, we locked in an itinerary in advance. I requested he optimize the route for driving.

One notable planning moment: Ferguson had plenty of accurate information about Tulsa’s Native American history, but I had to prompt him to start there. A local guide would probably start there instinctively, since it’s where Tulsa’s story begins. Ferguson had the facts, but not the instinct.

The locked itinerary worked much better – no more criss-crossing the city, no forgotten stops, etc. The downloadable Google Maps route worked cleanly. We ran everything on the phone this time, which worked better than juggling two voice-activated devices (which was like traveling with two people with hearing issues.)

Once the plan accounted for Native American history and the Tulsa Sound, the tour itinerary consisted of:

All good, but. What Ferguson still couldn’t do was admit when he didn’t know something.

Standing on Route 66 on 11th Street, we asked about the Cowboy Bob statue. Ferguson confidently described a “cowboy holding a rocket.”

Given a photo of the recently updated murals on the downtown Grooper building, Ferguson couldn’t identify any of the musicians pictured. But rather than saying “I don’t know” he responded every time with a confident word salad.

The Verdict: How Well Did ChatGPT Tour Tulsa?

Judged against the task of putting together a half-day tour of Tulsa for someone new in town, Ferguson held up reasonably well in round two.

Using ChatGPT or another LLM can be a decent way to get a quick taste of a city, enough to figure out what deserves a longer visit later.

While you can ask questions mid-tour and get on-the-fly answers, it’s not a substitute for an actual guide. Every fact it tells you will need a second look before you repeat it to someone else.

But for a fast first pass at an unfamiliar place, an AI chatbot tour wasn’t horrible. Not good enough that Tulsa’s current human-led tour providers should be worrying yet, however.

Photos from Our Tulsa Tour:

Lessons Learned

ChatGPT wasn’t good at ad hoc. Tell it to research deeply, build a set itinerary, and optimize the driving route.
ChatGPT wasn’t location-aware. Even after turning on location sharing in the ChatGPT app, we had to tell him where we were.
AI couldn’t always see. You can often upload photos or even a live video feed to AI tools and get real-time feedback, but in our case what we were looking at was too new. It also thought we could see hidden buildings three blocks away.
AI forgets. Ferguson forgot where we had been and then lost an entire tour transcript with no explanation.
AI will make odd choices. I never got a satisfactory answer for why Ferguson sent us to that not-interesting section of Route 66.
AI is often confidently incorrect or vague. I had to push back a few times before Ferguson would admit to not knowing something.
AI will push boundaries. Ferguson was anxious to extend his role past the initial prompt.