Two years ago, I decided to take better care of myself and (re)start long-distance running. I trained to run half-marathon races again, and then I quickly realized marathon races were within reach. I’ve now run several half-marathon races, and that distance is now part of my regular weekend running schedule. I’ve completed three marathons, and I’m not going to stop.

I could always run without entering races, but there’s a great side-effect of joining competitions that I happen to love: the travel that goes with it. With marathon races, I set aside a few days of recovery right after the race, which I use to go sightseeing.

I’m telling you all that because I’ve just raced the Tokyo marathon, which was incredible and a little harder than I had planned. But I’m not about to tell you what went right and wrong and what sort of tech I used. I did that with previous races.

I will show you how I used ChatGPT Deep Research to visit Tokyo the week after the race and make the most of it. Now that I have used OpenAI’s AI agent as a travel assistant, I realize how amazing it is to have AI at your fingertips to help you plan your next trip that way.

I used ChatGPT to prepare for the trip long before I actually got Deep Research access. The feature launched only a few weeks ago and was exclusive to ChatGPT Pro users. That’s the $200/month ChatGPT subscription, which I can’t justify for my current needs. I waited for OpenAI to make it available to the ChatGPT Plus tier, which costs a tenth of that.

When a friend asked what I would visit in Japan, I said I already had chats with ChatGPT about it, but I didn’t have a specific plan. It all depended on how the marathon race would go and how I would feel in the days following it.

At the time, I had no idea that Deep Research would become available during my stay in Tokyo. But I knew I would integrate museum visits into my trip, as that’s something I always do when I travel.

My first prompt to ChatGPT (GPT-4o) was extremely simple. I asked for a list of the best museums in Tokyo. I wanted all the details about each of them, including their size. In the days following marathon races, I’ll reduce my walking significantly to give my legs time to recover, so my priority was visiting places that wouldn’t require that much walking.

As you can see above, ChatGPT gave me the answers I needed. It listed 20 suggestions, even though I didn’t ask for a specific number.

Before generative AI programs like ChatGPT were a thing, I’d have tackled this problem by buying a guide (in print or app) and looking at travel blogs for suggestions. But now that ChatGPT exists, I ignored all of that and trusted the AI to surface the information I needed.

I was working with this chat in Tokyo when Deep Research became available to my account. Again, I’m a ChatGPT Plus user in Europe, so I had to wait a while for OpenAI to bring Deep Research to my neck of the woods.

Speaking of location, ChatGPT isn’t available in Japan. Use a VPN while you’re there, and you won’t have a problem accessing the AI.

When the Deep Research button appeared in the ChatGPT composer, I changed my strategy. I started a different chat and gave the AI a more detailed prompt.

I asked ChatGPT to provide me with a list of 20 museums to visit during a 7-day trip. Walking distance was a priority, but I also told the AI to consider local travel passes that would give me free or cheaper access to these museums. I also told ChatGPT to include attractions that aren’t covered by passes so as not to miss other exciting options.

Deep Research went to work, but only after asking a few additional questions. ChatGPT wanted to know what sort of museums I preferred, whether I wanted interactive exhibits, and whether the travel pass should cover public transit in addition to museum fees. I answered the questions, and Deep Research went to work while I prepared for my day.

The research took 17 minutes to complete, with ChatGPT inspecting 52 sources to give me a detailed Tokyo museum guide that met my needs.

The resulting report was amazing and overwhelming at the same time. I reviewed the recommendations over breakfast, trying to figure out how to put them to good use.

Guess what you should do once ChatGPT compiles such a travel guide for you? Ask more questions. That’s what I did. First, I told the AI to arrange the museums by pass type. Then, I started asking questions about locations. I also wanted to be as efficient as possible when moving from one place to another.

For example, I wanted to visit the Imperial Gardens, so I told the AI to give me the closest museums ranked by descending closing time from the Deep Research list.

I also asked ChatGPT to arrange the museums in proximity to each other so I could figure out daily itineraries. Again, I wanted to be as efficient as possible and make the most of each day. I didn’t want to walk any more than necessary between locations, especially in a massive city like Tokyo.

I could have told the ChatGPT to create itineraries for me, but I didn’t want the AI to completely dictate how I would spend my days in Tokyo.

My point is that ChatGPT Deep Research can be an unexpected resource for all sorts of things. The best part is that Deep Research offers plenty of links, which I used to get tickets and information on some of the museums on the list.

ChatGPT can also continue the conversation about that Deep Research report for all your follow-up questions. That’s actually a key feature to take advantage of. The initial report should be only a starting point.

Think of it like being able to ask a travel guide or a blog specific questions about the information you’ve just read about the country or city you’re about to visit.

That’s just one way to use Deep Research, and I’ve only started exploring this ChatGPT AI agent. Turning ChatGPT into an improvised travel agent is something I’ll keep in mind for all my future running trips.

By the way, ChatGPT is one of the reasons I’ve been able to run so many half-marathons and marathons in the past two years. I used AI to come up with training plans long before AI agents were a thing.