How to use AI to plan a trip: 6 prompts, and the 5 AI tools that help you book

Which AI can search, read maps and book a trip?

Five AI tools now handle part of a booking, and in every one a hotel, airline or travel site holds the reservation, not the AI company. Google says it plainly for AI Mode: "the hotel or booking platform will act as the merchant of record." That's who you call when a plan changes. Each row below comes from the vendor's own page, read on 12 October 2026, except Perplexity's, which comes from a March 2025 Skift test and Perplexity's own August 2025 changelog.

Gemini is the pick for a plan built on real places, because Google wires it to Maps, Flights and Hotels. Google's Gemini trip planning post says it pulls locations, reviews and prices from those three services. ChatGPT and Claude both reach Booking.com on a free account. OpenAI's search help page puts web search on the Free plan, and the Expedia and Booking.com apps launched for Free users too. Claude searches with a citation on every answer, and since April 2026 its connectors reach Booking.com, Tripadvisor and Viator on all plans.

Dedicated planner apps do a narrower job. Wanderlog puts your itinerary and your map on one screen and imports bookings you forward from email. Mindtrip says it can "find and book flights, stays and restaurants." They aren't immune to the same mistakes, though: the BBC reported in 2024 that the Layla app briefly claimed there was an Eiffel Tower in Beijing.

Can AI help you choose where to go?

AI is good at choosing a destination once you give it limits instead of a wish. "Somewhere warm in March" gets you the same ten places every travel magazine lists. Add a budget, a flight time and what you'd hate, and ChatGPT, Gemini or Claude can narrow it to three you'd actually book:

Destination prompt: "Suggest three places for [number] days in [month], flying from [city] with at most [hours] of flying each way. Budget: [amount] per person, flights included. We like [interests] and dislike [crowds, long drives, heat]. For each place, give the usual weather that month, one reason it fits us, one reason it might not, and the entry documents a [country] passport holder should check."

Then ask it to argue against its own top pick. The weak spot it names, like a rainy season or a festival that fills every hotel, is the thing you check first.

What prompt gets a usable trip itinerary?

A usable itinerary prompt gives the model your arrival and departure times, where you're sleeping and how fast you move. Leave those out and ChatGPT or Claude falls back on a generic three sights a day. Fill the brackets:

Itinerary prompt: "Plan [number] days in [city] for [who is traveling]. We land at [time] on [day] and leave at [time] on [day]. We're staying in [neighborhood]. Pace: [two things a day, or packed]. We love [interests] and want to skip [things]. Food and activities budget: [amount] a day. Group each day by neighborhood so we don't cross the city twice, add walking or transit time between stops, and give each day one indoor backup. After the plan, list every opening hour, ticket price and last entry time you used, as a checklist I can verify."

The checklist at the end is the part that saves the trip. It turns every fact the model might have wrong into a line you can tick off. Then push back with three follow up questions:

• "Which of these places need a timed ticket or a booking in advance?"
• "Move anything that's closed on [day of week] and tell me what you moved."
• "What did you assume that I didn't tell you?"

The last one finds the hidden defaults, like a rental car you don't have or a 7am start with small children. The travel prompt pack has 20 more prompts built the same way, from a work trip with one free day to a road trip.

How do you plan a trip budget with AI?

AI builds a good budget from prices you give it and a bad one from prices it remembers. Every model stops learning at a knowledge cutoff, so a remembered hotel rate is a guess with a date you can't see. Paste real prices instead:

Budget prompt: "Build a budget for [trip, dates, travelers] in [currency]. Split it into flights, lodging, local transport, food, activities and a 10% buffer. Use these prices I found: [paste fares, room rates and ticket prices]. Where I haven't given a price, give a low and a high estimate and label it as an estimate. Don't invent fares. End with the total per person and the three costs most likely to change."

Turn web search on and add "search for current prices" when you want the model to fill the gaps itself. A search result still isn't a bookable price. A fare found on Tuesday is a lead. The checkout page is the price.

Can ChatGPT check visa requirements?

ChatGPT can list the documents you might need, but only a government site can confirm them, because entry rules change faster than models are retrained. The UK changed its rules for visitors three times in about a year. Its electronic travel authorisation (ETA) rose from £10 to £16 on 9 April 2025, per the Home Office fee table. On 25 February 2026, the UK began stopping visitors of 85 nationalities from boarding without one. On 8 April 2026 the fee rose again, to £20, per the April 2026 fee table, and that's the price GOV.UK lists today.

The EU moved too. Its Entry/Exit System began on 12 October 2025 and replaced passport stamps on 10 April 2026, recording fingerprints and a facial image for non EU visitors on short stays, per the European Commission. ETIAS, the EU's planned travel authorisation, hasn't started. The EU's ETIAS page says "ETIAS is currently not in operation" and that it collects no applications yet. A site charging you for ETIAS today isn't the EU.

A fresh model doesn't solve this. Anthropic lists a June 2026 knowledge cutoff for its newest Claude models, so any rule that changes after that month reaches Claude only through web search. Use AI to build the list of questions, then answer them from official pages:

Entry rules prompt: "I hold a [country] passport [and a residence permit from country, if any]. I'm flying [full route, including layovers] on [dates] for [tourism or business]. For each country on the route, including transit stops, list the entry documents I might need and the official government page where I confirm each one. Don't state fees or processing times as facts. Mark them as things to check."

Then check each answer at the source. USA.gov sends US citizens to the State Department's country pages on travel.state.gov, which list entry, exit and visa requirements. British travelers have GOV.UK's foreign travel advice. The destination's own government has the final word, and layovers count: GOV.UK says a traveler on a connecting flight who goes through UK passport control needs an ETA too.

What does AI get wrong when planning a trip?

AI gets opening times, seasonal closures and sometimes whole places wrong, in the same confident tone as everything else. The BBC collected cases in September 2025. On the Japanese island of Itsukushima, Dana Yao and her husband timed a sunset hike up Mount Misen around a ChatGPT plan. ChatGPT told them the last ropeway down left at 17:30. It had already closed, and they were stuck at the summit.

In Peru, a tourist paid nearly $160 to reach a rural road near Mollepata, heading for a "Sacred Canyon of Humantay" that a local guide says doesn't exist. The same BBC report cites a 2024 Sainsbury's Bank survey: 37% of people who used AI to plan travel said it couldn't give enough information, and about 33% said its recommendations included false information.

The fix is a second pass with web search on, aimed at the facts that move. A model that searches can still read an old page, so ask for each source and its date:

Fact check prompt: "Search the web for each place on this list. For each one, give the official website, the opening hours it lists for [dates], any seasonal closure, and the date that page was last updated. If you can't find an official page, say so instead of guessing. [paste list]"

Open the venue's own site for anything with a last entry or a last departure: ropeways, ferries and museums with timed tickets. Our guide to AI hallucinations explains why a model fills a gap with a confident guess.

Should you book a trip through ChatGPT?

Book through ChatGPT only if you'd book with Expedia or Booking.com directly, because they're the companies holding your trip. OpenAI launched both apps in ChatGPT on 6 October 2025, for signed in users on Free, Go, Plus and Pro outside the European Economic Area, Switzerland and the UK. Its launch post describes searching inside the chat and lists instant checkout, through its Agentic Commerce Protocol, as coming later.

Google AI Mode goes furthest. Since 27 August 2026 it books US hotels in English through 10 partners, Booking.com, Expedia, Hilton and Marriott among them, and you pay with Google Pay, per Google's announcement. Flights are different: Google says you still buy the ticket on the airline's website or a booking platform. Claude's connectors book through the partner too, Booking.com or Resy for example, and Anthropic says Claude checks with you before it books or buys anything.

Whichever route you take, read the cancellation terms on the confirmation and keep the travel company's booking number. A chat transcript isn't a receipt.

How do you make a packing list with AI?

Give the model the forecast, not the month. "Lisbon in April" gets you an average, and your week might be the wet one.

Packing prompt: "Make a packing list for [number] days in [places] from [dates]. Forecast: [paste the highs, lows and rain chance for each day]. We're doing [activities] and traveling with [cabin bags only, or checked bags]. Laundry: [yes or no]. Group the list by bag, and flag anything with airline or customs limits, like liquids, power banks and food."

Whizi, which publishes this page, fills in that forecast line for you on its website. Ask about the weather and Whizi pulls up to 8 days of forecast from MET Norway before the model answers, then draws it as a weather card, so the highs and lows come from a forecast and not from memory. Web search is on by default in every new chat. It loses on the rest of the trip: Whizi shows no maps and books nothing, so the table at the top still decides where you book.