This summer, we traveled through Europe by train and got a personalized podcast to help plan our days in each new place.
One Friday evening in St. Moritz gave us the only episode of Alpine Evenings that will ever be produced. 🚂🏔️
Forget guidebooks or top-10 lists. You probably already use an LLM to plan your travels anyway.
In addition to the location, we gave the model more context: what time we arrived, how long we’d stay, our energy levels, where we were staying, and what kinds of activities we wanted to avoid or include. That resulted in a solid starting point.
But the real value came from turning the suggestions into something easier to share and discuss. We asked the model to role-play as a podcast host and write a short, personal episode script.
In about two minutes, we could listen to our brand-new podcast together using the text-to-voice feature.
Quick to consume, easy to react to, and often surprisingly accurate. Some suggestions were so on point they completely changed what we ended up doing.
Since LLM outputs are inherently stochastic, each show got its own name, tone, and style depending on the day's prompt and context.
Research and voiceover worked best with Gemini 2.5 Pro. It delivered both grounded suggestions through Google search and one of the best voice models we tried.
Is this a product idea? Possibly.
Real-time, personalized podcast episodes for specific travel contexts. But it’s important to call out the risks: the model sometimes gave incorrect suggestions or completely missed the point. Used blindly, that could be frustrating, or even unsafe. It helped a lot that I knew exactly what went into the prompt and could reason about the output.
Still, if you understand the limitations, I really recommend trying this approach manually. Ask your favorite LLM for a travel recommendation tailored to your situation. Then convert it into a podcast script and have it read aloud.
It turns planning into something social, conversational, and fun. And it definitely improved our trip.
Traveling Europe guided by our personal AI podcast was first published 2025‑07‑05