The biggest mistake people make with AI travel planning is being too vague. "Plan my Paris trip" produces a generic tourist checklist. The approach that actually works is treating Claude like a travel agent who needs real details to do their job. Here's the full system.
Be specific, not generic
The more specific the input, the better the output. A strong itinerary prompt includes: destination, exact dates, travel style, budget per day, physical limits (how much walking), and specific interests. Not just "I like food" but what kind of food, whether you want cooking classes, market visits, or just the best neighborhood spots.
Compare these two prompts. The first gets a list of Paris clichés. The second gets something actually useful:
Weak: "Make me a Paris itinerary for 5 days."
Strong: "5 days in Paris in late September, walking 8–10 miles/day is fine, interested in Haussmannian architecture and hidden courtyards, prefer neighborhood restaurants over tourist spots, budget €80/day for food, already have Louvre tickets for day 2. Give me a day-by-day breakdown with morning/afternoon/evening blocks, best transit between areas, realistic timing, and 2–3 restaurant options per meal with addresses."
Enable web search in Claude
Claude with web search enabled can find current information — updated restaurant hours, recent reviews, current ticket prices, and transit updates. In Claude.ai, this is the "Search" tool. Turn it on before submitting your itinerary prompt. Without it, Claude is working from training data which may be months or years old.
Ask for mobile-friendly format
Add this to your prompt: "Format this for mobile use — each day on its own section, include Google Maps addresses, walking distances between stops, and realistic timing estimates. Flag anything that requires advance booking." This makes the output copy-paste ready into Apple Notes or Notion for offline access.
Request local customs and practical context
Ask Claude to include a "local customs" section: tipping culture, restaurant timing norms (when do locals actually eat dinner?), transit etiquette, neighborhood safety notes, and anything a first-timer would learn the hard way. This context makes the itinerary actually usable, not just a list of places.
Follow up and refine
The first output is a draft. Iterate: "Swap Tuesday afternoon for a flea market visit instead" or "The restaurant you recommended is fully booked — what's the best alternative in the same neighborhood?" Claude remembers your constraints and adjusts without starting over.
Here's the full example prompt that produces the best results:
Save the Claude conversation. Every follow-up question you ask makes the itinerary smarter — and if you need to swap something out, Claude already knows all your constraints. Bookmarking the thread beats starting over every time.