Itineraries guide · Trip planning
Can you trust an AI travel itinerary? How to check one in 10 minutes
AI is good at a first draft and bad at facts that change. Here’s what the 2026 surveys found, the mistakes that strand people, and a 10-minute check for any plan.

Short answer. Trust an AI itinerary as a first draft, not as a plan. In a 2026 survey of US travellers, 55% of those who used AI had received at least one wrong recommendation, and 16% had been sent to something that didn’t exist. Before you rely on it, check that every place exists and is open on your day, that the journeys between stops fit the time, and that entry rules come from an official source.
AI is very good at the first hour of trip planning. Ask for three days in Lisbon with a food focus and you’ll get a tidy, confident plan in seconds, with neighbourhoods grouped sensibly and a pastry stop every morning. The trouble starts when you treat that plan as checked. It isn’t. It was written from memory, and memory doesn’t know that the café closed in March or that the museum shuts on Mondays.
So can you trust an AI travel itinerary? As a draft, yes. As the plan you’ll follow on the day, only after you’ve checked it, and that takes about ten minutes.
How often are AI travel itineraries wrong?
Often enough that it’s normal, not unlucky. The best recent numbers:
- Greetwell’s AI Travel Report 2026 surveyed 1,000 US leisure travellers in August 2026; 485 of them used AI for travel. Of those, 55% had received at least one recommendation that was wrong: inaccurate, unavailable, closed or non-existent. Breaking that down, 27% got incorrect prices, hours or other details, 21% found the place fully booked or unavailable, 16% were recommended something that didn’t exist, and 10% were sent to a business that had closed for good.
- A Censuswide survey for Qlik of 2,001 US adults in June 2026 found that 58% use AI tools to plan trips, and a quarter named inaccurate or outdated information as a frustration. 15% said AI mistakes had significantly disrupted or ruined past plans.
- Which?, the UK consumer group, tested six AI tools on 40 consumer questions in September 2025. ChatGPT wrongly said travel insurance is mandatory for visits to the Schengen area, and Copilot wrongly said you’re always entitled to a full refund for a cancelled or delayed flight.
Even OpenAI’s own help centre says ChatGPT can sound convincing while giving "incorrect or misleading information", and advises users not to use it as their only source.
Why does AI get travel facts wrong?
Three reasons, and they explain which parts of a plan to check:
- It writes from memory, not from today. A language model learns from text up to a point in time. Opening hours, prices, closures and rules that changed since then are simply unknown to it, and it won’t say so.
- It fills gaps with what sounds likely. When a model doesn’t know, it can produce a plausible name or time instead of admitting it. That’s how a non-existent canyon or a last cable car at 17:30 appears.
- It doesn’t measure the map. A plan can list two "nearby" places that are 40 minutes apart on public transport, or a day trip where you’d spend longer travelling than being there.
None of this makes AI useless. It’s good at structure, ideas and wording. It’s bad at the facts that change and at distances it never measured.
What happens when nobody checks?
These cases were all reported in 2025 and 2026:
- A canyon that doesn’t exist. Travellers in Peru set out for the "Sacred Canyon of Humantay", an AI invention, paying about $160 to reach a rural road before a local guide stopped them. The case was reported by BBC Travel and is logged in the AI Incident Database.
- Stranded on a mountain. On Miyajima in Japan, a couple used ChatGPT’s time for the last ropeway down Mount Misen. The ropeway had already closed when they reached the top. Rick Steves' team wrote about the same story.
- Hot springs that weren’t there. In Tasmania, a tour company’s AI-written web page sent visitors to the "Weldborough Hot Springs", which don’t exist. The owner told reporters: "Our AI has messed up completely."
- A missing travel authorisation. A Spanish couple were refused boarding for Puerto Rico after ChatGPT said they didn’t need a visa but didn’t mention the ESTA, the electronic authorisation they did need.
Notice the Tasmania case: the error came from a website, not a chat. Itineraries copied from travel blogs and listicles need the same check.
How do you fact-check an AI travel itinerary?
Work through the plan once, in this order. For a three-day trip it takes about ten minutes.
| Step | What to check | How | Catches |
|---|---|---|---|
| 1. Every place exists | Search each named place on Google Maps | A real listing, with photos and recent reviews, in the right city. If it says "Permanently closed", strike it | Invented places, closed businesses |
| 2. Open on your day | Look at the opening hours for the weekday you’ll be there | Check closing days (many museums close one weekday), seasonal hours and public holidays at your destination | Closed doors, wrong hours |
| 3. Last departures | For cable cars, ferries, boats and the last train back, find the operator’s own timetable | Don’t trust a time from the plan; it’s exactly what AI gets wrong | Being stranded |
| 4. The journeys fit | Put each pair of stops into a maps app for the day and time you’ll travel | If the journey is longer than the gap between stops, the day doesn’t work | Impossible timing, cross-city zig-zags |
| 5. The day isn’t overloaded | Count the stops and add up the travel | More than 8 stops, or more than about 3 hours of travel in a day, is usually too much | Exhausting days |
| 6. Bookable things are bookable | For tours, timed entry and restaurants, open the official booking page | Check availability for your date and the cancellation policy | Sold-out or by-booking-only places |
| 7. Entry and rules | Visas, ESTA/ETA, transport passes, local rules | Only from official government sites, never from the AI | Being refused boarding |
If you only have two minutes, do steps 1 and 7. Those are the mistakes that cost the most.
For the pace limits in step 5, our guide on how to avoid an overcrowded itinerary explains how many stops a day is realistic.
What does a checked AI itinerary look like?
Here’s an honest example from our own app. On 21 October 2026, Amicly’s AI wrote a three-day food plan for Lisbon, and then the checks ran on it. Of 18 places, 16 matched a Google Maps listing, and 11 journeys between stops were timed.
The two that didn’t match were the National Pantheon (Church of Santa Engrácia) and the Miradouro da Senhora do Monte viewpoint. Both are real and well known. The check couldn’t match the names closely enough to Google’s listings, so it labelled them "unverified" rather than guessing. That’s what you want a check to do: tell you which two to look at yourself, not quietly pass them.
One more thing the check could do nothing about: no travel dates were entered, so it couldn’t test closing days. A plan with dates gets more checks than one without.
When can you trust AI travel answers without checking?
For things that don’t change and don’t cost much if they’re wrong:
- Ideas for what a city is known for, and which neighbourhoods suit what
- A rough shape for the trip: how many days where, what to group together
- Packing lists, phrases to learn and general etiquette
- Rewording your own notes into a cleaner plan
Check anything with a time, a price, an opening day, a distance, a booking or a rule. Which?'s advice is the same: ask the AI for its sources, check them yourself, and get a second opinion when the stakes are high.
A faster way to check a plan
You can do all of the above by hand with Google Maps and a coffee. Or paste the itinerary into Amicly’s "Check my itinerary". It looks up each place on Google Maps, times the journeys between stops, runs the closing-day, pace and doubling-back checks in code, and lists problems first, each with its evidence and a suggested fix. Places it can’t match are listed as unverified, never guessed. Your first check is free.
If you’re building a plan from scratch rather than checking one, start with how to make a travel itinerary.
Sources
- Greetwell: AI Travel Report 2026 (September 2026) · checked 29 October 2026
- Qlik via Business Wire: AI has become the preferred vacation planning tool (July 2026) · checked 29 October 2026
- Which?: ChatGPT and Gemini among AI tools giving risky consumer advice (November 2025) · checked 29 October 2026
- AI Incident Database: Incident 1636 (non-existent "Sacred Canyon of Humantay", Peru) · checked 29 October 2026
- Rick Steves' Europe: AI for trip planning? Tread with caution (January 2026) · checked 29 October 2026
- Euronews: How an AI-generated travel website sent tourists to a non-existent hot spring (January 2026) · checked 29 October 2026
- Gulf News: ChatGPT travel tip strands Spanish couple at airport (August 2025) · checked 29 October 2026
- OpenAI Help Center: Does ChatGPT tell the truth? · checked 29 October 2026
- Wikimedia Commons: Tourists reading, Largo Camões, Lisbon (CC BY 2.0) · checked 29 October 2026
- Google Maps Help: Download areas and navigate offline · checked 29 October 2026
Spotted something out of date? Report a correction. How we review