Travel Itinerary Planner is a Claude AI skill — live-researched, calibrated to your travel style, day-by-day.
Ask five different people to use generic AI to plan a trip to Barcelona. Give them all different prompting styles — one who writes detailed briefs, one who types three words, one who specifies food preferences, one who mentions they're travelling with a toddler. The resulting itineraries will be recognisably similar: La Sagrada Família, Gothic Quarter, La Boqueria, Barceloneta beach, Montjuïc. Maybe Park Güell if they're thorough.
This is the failure mode — and it's not the user's fault for prompting badly. It's what happens when AI plans from the median of everything it's read about a place rather than from an understanding of the trip you're actually taking.
Most people try to fix this with more detailed instructions. Add more constraints. Specify the neighbourhood. Tell it your pace. The output improves marginally. You're still essentially receiving a remix of the same canonical list, sequenced differently.
The Actual Problem: Optimised for Completeness, Not for Logistics
Generic AI produces itineraries that look thorough. Every major attraction is accounted for. The days are full. The structure is there — morning, afternoon, evening. When you read it on a laptop, it feels like a plan.
The problems emerge on the ground, and they're almost always the same category of problem: logistics that were never verified, timing that was never calibrated, and geography that was optimised for coverage rather than for actual movement through a city.
AI produces plans that pass the "looks right" test on a screen while failing the "works on arrival" test in the city — because it's never asked to verify the logistics, only to produce the plan.
Here's what that looks like across the three most common failure categories:
| What the plan says | What actually happens |
|---|---|
| "Visit the central market in the morning" | The market is closed on Sundays, which your trip starts on |
| "Quick walk between the cathedral and the old town" | It's 40 minutes and largely uphill — relevant if you're travelling with elderly parents |
| "Try the famous restaurant on Day 3" | The restaurant closed 18 months ago; it still appears in AI outputs because it was prominent in training data |
| "The museum is best visited in the afternoon" | The museum requires advance booking that fills up two weeks out; walk-ins aren't possible in peak season |
| "Day 2: Art museum + seafront + local neighbourhood" | These three locations are in different parts of the city; the sequencing adds 2+ hours of unnecessary transit |
None of these failures are about the plan being factually wrong — the market does exist, the museum is real, the walk is technically possible. They're about the plan being built from static knowledge rather than from current, verified information about a real trip.
That gap is exactly what the Travel Itinerary Planner skill for Claude was built to close.
Why More Prompting Doesn't Actually Fix This
The instinct when generic AI disappoints is to add more detail to the instruction. More context about your preferences, more constraints, more specificity about what you want. And this does improve output — up to a point.
Better prompting is trying to fix a knowledge problem with an instruction problem — and those aren't the same thing.
The issue isn't that the AI needs better instructions about your preferences. It's that no instruction you can write will tell it whether a specific restaurant is still open, whether the temple requires a ticket booked three days in advance, or whether November is when the city's main annual event crowds every accommodation within 5km of the centre.
That information isn't in the training data — or if it is, it's out of date. The model can produce a confident, well-formatted itinerary without knowing any of it. Confident formatting is not the same as accurate planning.
What a Skill That Researches First Changes
The Travel Itinerary Planner skill approaches the problem from the other direction. Before it writes a single day of your itinerary, it runs live research on your destination — current conditions, what's actually open, what's worth the trip at the time of year you're going, and where the known friction points are.
That research shapes what goes into the plan. Not just the attractions, but the sequencing, the timing, the logistics flags, and the advance booking notes that surface before they become a problem in the city.
The generic version lists everything. The NovaKit version plans for arrival-day tiredness, surfaces the specific booking windows that are realistic in October, sequences attractions by location rather than by fame, and flags the timing mistakes that people consistently make on the Colosseum visit.
The Calibration That Changes Everything
Beyond the research, the skill asks three questions before it starts — questions about your pace, your group, and what you actually care about. These aren't optional. They're the difference between a plan built for a fast-moving solo traveller who wants to cover ground and a plan built for two people who want to sit in cafés and wander without an agenda.
People planning trips where logistics matter: multi-city itineraries where order affects cost and sense, destinations with strong seasonal variation, trips with mixed groups (ages, mobility, interests), and bucket-list destinations you're unlikely to revisit and want to get right the first time.
The same skill, given different answers to those three questions, produces fundamentally different outputs for the same destination and trip length. A family with young children gets nap windows built into the afternoon. A couple celebrating an anniversary gets restaurant bookings flagged at the right moment in the planning process. A solo traveller who wants intensity gets a denser schedule than someone who explicitly says they don't want to feel rushed.
Generic AI can't do this from a single prompt — not because Claude isn't capable, but because no prompt contains enough context to make those judgements without asking. The skill earns the right to plan by understanding the trip first.
The result is an itinerary that holds up on arrival. Not perfect — you'll still deviate from it, discover things it didn't account for, make spontaneous choices. That's the point of travel. But you won't be standing in front of a closed museum on the first morning wondering whether the whole plan needs to be rebuilt from scratch.
The next piece most people tackle from here is a listing that converts browsers into confirmed bookings.
Put this to work: the Travel Itinerary Planner skill for Claude turns everything above into one guided workflow you run in a normal Claude chat. Not ready to buy? Start with a free Claude skill and see how it works first.