The first time you ask an AI to plan a five-day trip to Singapore, you’ll get back something that looks brilliant. Day-by-day itinerary, hotel within ten minutes of Marina Bay, restaurant suggestions matched to client meetings, even gym slots before breakfast. Then you check the flight number and it doesn’t exist. The hotel address is real but in a different city. The restaurant closed in 2019.
This is the specific failure mode every executive assistant runs into within their first week of using AI for travel. The model is brilliant at structure and confident on details. The details are sometimes wrong. The EAs whose principals never miss a connection have stopped trusting AI for facts and started using it for the parts of the job that don’t require facts at all.
Where AI is genuinely useful, and where it isn’t
AI is reliably good at three travel tasks: turning a vague brief into a structured itinerary, adapting an existing itinerary when something changes, and writing the briefing your principal will read on the way to the airport. It is unreliable for live flight prices, real-time hotel availability, restaurant operating hours, and visa requirements. The reason is simple: training data is months or years behind reality, and the model has no incentive to admit when it doesn’t know.
Use ChatGPT or Claude for the structuring work. Use a real booking interface (your corporate travel desk, Google Flights, the airline directly) for anything that involves actual money or a reservation number. The travel-specific AI tools that have emerged in the last eighteen months (Mindtrip and Wanderboat are the two worth knowing) sit between the two by pulling real-time data into a chat interface. They’re useful for inspiration and rough scoping. They are not a replacement for a corporate travel agent on a complicated multi-leg itinerary.
The brief that produces a usable first draft
Most AI travel planning fails at the input. EAs paste in two sentences (“five days in Singapore, executive trip, find me a hotel near Marina Bay”) and wonder why the output feels generic. The brief that actually works captures the constraints AI doesn’t know to ask about.
“You are helping an executive assistant plan a business trip. Your job is to produce a structured day-by-day itinerary that I will then verify against live flight and hotel data. Do not invent flight numbers, prices, or specific hotel availability. Trip details: [PRINCIPAL NAME] travelling from [DEPARTURE CITY] to [DESTINATION] on [DATES]. Purpose: [MEETING / CONFERENCE / SITE VISIT]. Confirmed meetings already in calendar: [LIST WITH ATTENDEES, TIMES, AND VENUES IF KNOWN]. Principal preferences: [AIRLINE / SEAT / HOTEL CHAIN / DIETARY / FITNESS / SLEEP-SCHEDULE NOTES]. Hard constraints: [VISA STATUS, BUDGET CAP, EARLIEST/LATEST FLIGHT TIMES, ANYTHING NON-NEGOTIABLE]. Output as a table with columns: date, time, activity, location, transport between, notes for principal. After the table, list every assumption you made that I need to verify before booking, and every detail I haven’t given you that would change the plan if I did. Mark anything that looks like a fact but might be out of date with [VERIFY].”
The instruction to flag assumptions and mark unverified facts is what turns AI from a hallucination machine into a useful first draft. You’ll get an itinerary plus a verification checklist. Work through the checklist before you book anything.
A counterintuitive observation: the best EAs deliberately give the model less authority than it wants. They will explicitly write “do not invent prices” and “do not assume restaurant hours” because they have learned the hard way that without those instructions the model fills gaps with plausible fiction.
Adapting itineraries when reality intervenes
The other place AI earns its keep is the moment a flight gets cancelled or a meeting moves. You don’t need the model to find a new flight (your travel desk will do that better and with real prices). You need it to rebuild the rest of the day around the new flight time, and to write the explanation your principal will read on their phone in the back of a taxi.
“My principal’s flight from [ORIGIN] to [DESTINATION] has been moved from [OLD TIME] to [NEW TIME]. Below is their original itinerary for the next 48 hours. Rebuild the schedule around the new flight time. Identify which scheduled meetings need to be moved and which can stay. Draft a short message I can send to each affected attendee. Flag any logistics that change (hotel check-in, ground transport, dinner reservations) and what I need to do for each one. Original itinerary: [PASTE]. New flight info: [PASTE].”
Run this and you have a triage plan in ninety seconds. Compare that to the half hour it takes to manually open every calendar invite and figure out the cascade. The model is not solving the underlying problem (the flight is still moved) but it is freeing you to be the human who makes the calls and reassures the people who need reassuring.
The pre-departure briefing your principal will actually read
The last piece of the workflow is the briefing your principal opens five minutes before boarding. It’s the document they will judge the entire trip on, because it’s the only one they read carefully. Generic templates produce skim-worthy briefings. A well-prompted AI produces ones they reference mid-meeting.
“Generate a one-page pre-trip briefing for my principal who is travelling to [DESTINATION] for [PURPOSE] from [DATE] to [DATE]. Sections: (1) The single most important meeting on this trip and why, in two sentences. (2) Each meeting in chronological order with attendees, the company’s recent news in one line, and the specific outcome we want from this meeting. (3) Cultural or contextual notes that will matter (local public holidays, election week, recent industry events). (4) Three open questions my principal should be ready to answer or ask. (5) Logistics they actually need to remember (passport expiry check, currency, plug type, embassy phone). Keep total length under 400 words. Use bullet points. Source material follows: [PASTE COMPANY DOSSIERS, MEETING NOTES, EMAILS].”
The model will not invent the meeting outcomes you want. You have to give it those, which forces you to think them through, which is the actual planning work. AI doesn’t replace the strategic thinking. It just makes sure none of the strategic thinking goes uncaptured.
What stays human
The travel piece of an EA’s job is one of the most exposed to AI on paper, and one of the most resistant in practice. The reason is that the value isn’t in the itinerary table. It’s in the moment your principal lands at 11 p.m. local time, exhausted, and the car is exactly where you said it would be. AI helps you get the table written faster. The car still gets there because you, a person, called the driver yesterday to confirm. The principals who keep the same EA for a decade are not paying for the table.