I planned my entire vacation with Claude, and I'm never going back

I’m not a big fan of LLMs as decision-making tools, but they do serve a purpose in my workflow as an organization layer. Specifically, when it comes to travel. The problem is that if you ask Gemini, ChatGPT, or Claude to plan a multi-day trip to Tokyo, you will probably get a generic list of tourist attractions. It might tell you to visit Tsukiji market at noon, suggest a restaurant that closed two years ago, and send you across the city three times in one afternoon. That tracks, since it’s just basing its recommendations on the collective knowledge of the internet — both old and new.

AI doesn’t really know your preferences, how much walking you can handle, or whether visiting a particular spot actually makes sense given everything else on your itinerary. I figured this out earlier this year while planning a three-week trip to Japan.

My approach was a bit different. Instead of asking Claude to be my travel planner, I treated it more like a tool to manage logistics. I gave it all the information I had collected, my preferences, and the things I wanted to see, and then let it figure out how everything fit together. The result was the smoothest vacation I have ever planned. More importantly, I spent far less time jumping between Google Maps, spreadsheets, blog posts, and booking websites.

The trick is to let AI organize, not decide.

If this entire exercise has taught me anything, it is that the biggest mistake you can make with AI travel planning is giving it complete control. The generated itinerary might look okay on paper, but as someone who was already familiar with Japan because of prior trips, I could immediately see how superficial and touristy the recommendations were.

Claude does not know that I would rather spend an afternoon browsing vintage camera shops than visit another department store. It does not know that I would rather wake up at 5 a.m. for a good photo than sleep in. And it certainly does not know how tired I will be after several days of walking around Tokyo.

So, I changed the workflow.

Instead of leaving it to Claude, I curated all the things I actually wanted to do first. My curation went broad, including everything from Instagram posts to YouTube recommendations, blog posts, restaurants, shops, photography locations, and places I had discovered while researching Japan. Claude’s job was to turn that giant pile of information into something usable.

I created a dedicated Claude Project for the trip and uploaded a master document containing everything it needed to know. At the top of that document, I kept a simple decision log. Every major decision went in there, such as dropping a side trip, changing our hotel, or deciding how many days to spend in a city.

Maintaining a master document with constantly updated information on the decisions I’d made meant I could start a new conversation within the project without having to explain the entire trip again.

I gave Claude a lot of context

As with any AI tool, the more specific you get, the better the results. Claude is no different.

I even used the Chrome MCP to give Claude access to live web pages. When I found an interesting blog post about a photography location, I could give Claude the URL and have it pull out the useful information.

Instead of my reading through a 2,000-word blog post looking for a train schedule or admission fee, Claude could extract the relevant details and figure out whether the location actually fit into our plans.

I also created a few personal reference documents. For example, I have a severe seafood allergy, so I created a guide covering safe Japanese dishes, ingredients I needed to avoid, and useful phrases for explaining the allergy at restaurants. I also had a list of camera lenses I wanted to take, vintage watch stores I wanted to visit, and Japanese streetwear brands I wanted to browse.

This made a huge difference when planning individual days. At one point, I dumped a long list of local designer brands into Claude. Instead of giving me another giant list, it clustered the stores into areas that made sense geographically. I could then turn those locations into a Google My Maps file.

Something that would have taken me hours of opening addresses and moving pins around took minutes. The key takeaway for me has been that you can’t remove the human curation element of travel planning. In fact, Claude — and I suspect any AI agent — operates better with better filtering, and there’s no better filtering than human taste.

I stopped planning every minute of every day

The biggest change with using Claude as the logical reasoning layer was in how I structured the itinerary. For our three weeks across Tokyo, Kyoto, and a number of small villages, I stopped trying to create a rigid schedule for every hour.

Instead, I divided the itinerary into fixed anchors and flexible options. The anchors were things that absolutely had to happen. That could be a pre-dawn photography session at Fushimi Inari or an evening event that we had already booked.

Everything else went into flexible pools. These included cafes, restaurants, museums, shops, neighborhoods, and other things we could do if we happened to be nearby and had the time or energy.

This made the itinerary much more realistic. Nor did it require me to maintain a mental checklist. Claude was particularly useful for the boring logistical stuff. It could look at sunrise times, train schedules, opening hours, and the locations of different attractions, and work out whether everything actually fits together.

It also caught small details that I might have missed. For one photography morning, it worked out which train we needed to take to reach the Arashiyama bamboo grove before the crowds arrived. It also flagged closing days for smaller museums and reminded us about a weekend-only train we needed to account for.

When booking the Shinkansen from Tokyo to Kyoto, it even reminded us which seats to choose for a better view of Mount Fuji.

None of these things are particularly difficult to figure out. The problem is that there are dozens of such small decisions to be made on a multi-week trip. I’ve traveled to well over 100 countries, and even I can’t keep track of everything in my head. Having Claude keep track of them meant I didn’t have to.

The real benefit came when things went wrong

Even the most well-planned travel itinerary starts falling apart once you’re on the ground. So, it’s understandable that the biggest test came once we were actually in Japan. You get tired, it rains, a train is delayed, or you simply decide you don’t want to walk another 15,000 steps that day. Perhaps you’re lugging around too much shopping. Don’t ask me where that came from.

On our first afternoon in Shinjuku, I could tell we were too tired for the ambitious plan we had originally made. I told Claude where we were and asked it to find some low-energy options nearby.

Because it already knew our preferences and the rest of our itinerary, it could suggest things that actually made sense. It also reminded me about a nearby ramen restaurant that could accommodate my seafood allergy.

Another morning, heavy rain completely changed our plans around Mount Fuji. Instead of opening Google Maps and starting from scratch, I asked Claude to pivot the day. It pulled options from the flexible list we had already created and suggested nearby indoor activities, shrines that would still be interesting in the rain, and places to eat.

That’s where the system starts working in your favor. No matter how good a traveler you are, you don’t really have innate spatial awareness on the ground. Moreover, if you’re anything like my partner, the deluge of recommendations for places to visit on social media makes it impossible to keep track of everything. When my plan started falling apart, I didn’t have to figure things out by myself; I just asked Claude to pivot to my personally curated fallback options.

AI as the travel organizer, not the travel agent

The interesting thing about this approach is that Claude did not actually decide what I wanted to do. I did that part. Depending on where you lie on the spectrum of AI love, you’ll either love or hate that. For me, human curation remains the way to go.

I chose the places, restaurants, shops, photography spots, and experiences that interested me. Claude handles the annoying part of connecting everything together. Between tracking opening hours, train schedules, and all the little details that add tedium to travel planning, Claude or any other AI tool can be a very useful companion.

What none of these AI tools do very well is the part that needs taste, expertise, and curation. But I shouldn’t have to remind you that AI is no replacement for human curiosity.

I’ve since tried this approach on another, shorter trip, and it works just as well. For me, AI isn’t the tool that decides what I should do in a city. Instead, I tell it everything I want to do, and it helps me decide when I should do it, how I should do it, and when the plans fall apart, what I should do instead.

That small change in approach made a huge difference to my trip, and I don’t think I would go back to planning a vacation the old way.