
Travel planning has become one of the most popular uses for AI assistants. To put that claim to the test, we asked an assistant to put together a three-day Matera itinerary that took our budget, walking distances and food preferences into account. Then we followed the route without changing a thing. Its suggestions for historic buildings and viewpoints were spot on, but its information on opening hours and some restaurants was out of date.
In this article
- How we wrote the prompt
- Suggestions that held up on the ground
- Where the assistant got it wrong
How we wrote the prompt
When writing the prompt, our first rule was to reduce ambiguity. We told the assistant: three days, two people, a mid-range budget, no more than six kilometers of walking a day and a travel companion who doesn't like spicy food. We also mentioned that we didn't want early mornings and that we love taking photos. The more detail we gave, the more personal the route became.
On the first try, the assistant produced a generic tourist list. When we reworded the prompt to “give separate suggestions for morning, afternoon and evening each day, with an estimated time for each,” the result improved markedly. In other words, talking to an assistant is a skill too: stating a clear goal, limits and the format you want up front is far more efficient than fixing things afterward.

The assistant drew up a great route, but it had no way of knowing the door was locked.
Mark Ellison
Suggestions that held up on the ground
On the ground, the most accurate suggestions were historic buildings and viewpoints. The valley view from the terrace the assistant suggested we visit at sunset gave us the best shot of the trip. The plan to stroll the old quarter during quiet hours and leave the main street for the back alleys also worked. The walking route focused on stonework was a real pleasure and never felt rushed.
Some of the food suggestions were spot on too. The assistant correctly recommended two restaurants with mild options and pointed us to a small family-run shop for dessert. The route, which stayed within budget and matched our walking limit, showed that the software could take user preferences seriously. But not everything went so smoothly; the real problems surfaced in details about how current the information was.
Where the assistant got it wrong
The assistant's biggest weakness was up-to-date information. The museum it suggested for the second afternoon was closed that day for maintenance. One of the recommended restaurants had moved. Because the assistant presented this information so confidently, we would have wasted time if we hadn't checked. Opening hours and reservation availability should always be confirmed through the venue's own channels.
| Topic | AI |
|---|---|
| Highlights | artificial intelligence, app, guide |
| Reading time | 8 minutes |
| Written by | Mark Ellison |
Key takeaways
- The more details we gave, such as time, budget and walking distance, the better the route.
- Suggestions for historic sites, views and food largely held up on the ground.
- Opening hours and addresses were out of date, so always verify them separately.
In the end, using AI to draft a travel plan is very efficient, but treating it as the final word is risky. Checking suggested places in a map app, looking at recent reviews and confirming opening hours separately all make for a better trip. The assistant built the skeleton of the route; we filled in the details.
What do you think about this post?


































































Leave a comment