1 in 6 AI trip planners got sent to a place that doesn't exist, survey finds
Key Highlights
- 1
- 2
- 3
+2 more in the takeaways below
One in six travelers who used AI to plan a trip were sent to a place, tour, or activity that doesn't exist. That's the new number from a Greetwell survey this week, and 55% of AI users hit at least one wrong recommendation.
That's what researchers call a hallucination, and it's the sharp edge of a bigger habit: more people trust AI with more of their trip every year. Here's what the survey found and what to check before booking.
- The core number: 1 in 6 AI-using travelers were sent to a place, tour, or activity that doesn't exist, per a new Greetwell survey.
- It's common, not rare: 55% of AI trip planners hit at least one wrong recommendation, from bad hours to a fully booked venue.
- Trust drops fast: a bad AI recommendation made travelers about four times more likely to say their trust in AI fell.
- The fix is boring: cross-check any AI-suggested place on a live map before building a day around it.
- This isn't new: documented cases of AI inventing destinations go back to at least 2023.
What the survey found
16% of AI-using travelers were sent to a place, tour, or activity that didn't exist, and 55% hit at least one bad recommendation of any kind. That's the headline from Greetwell's August 2026 survey of 1,000 U.S. adults, 485 of whom had used or tried AI for travel.
Greetwell is a travel concierge company, not a government agency, and this is one commissioned survey of 485 AI users, not a census. Worth saying, since the number is getting repeated everywhere without that caveat. It's a Stacker-produced wire piece that ran unedited on a dozen-plus TV-affiliate sites this week, including WSB-TV Atlanta, and none of those copies says what to check before booking. That's the gap this fills.
Who it actually affects
It hits anyone who used AI to plan a trip. Per the Greetwell survey, 53% of travelers had used AI for trip planning: 34% for research only, 14% for both planning and booking, and 5% who tried it and didn't find it useful. The 16% and 55% figures apply to that group, not to travelers overall, and that group keeps growing. Deloitte's own research puts gen-AI use for trip planning at nearly a quarter of travelers in late 2025, roughly three times the 2022 rate.
The failure types aren't evenly split. Here's the breakdown among AI users who hit any problem:
| Failure type | Share affected |
|---|---|
| Incorrect price, hours, or details | 27 |
| Fully booked or unavailable | 21 |
| Wrong for what was asked | 17 |
| Place, tour, or activity didn't exist | 16 |
| Permanently closed business | 10 |
Those five add up to well over 55% because they overlap: one trip can rack up more than one failure. The most common issue isn't the invented destination. It's a real place with wrong details attached, and in some ways that's worse. A fake cable car is obviously fake on arrival, but a wrong opening time just quietly ruins your afternoon.
The trust hit is lopsided. Among travelers who hit a bad recommendation, 21% said their trust in AI declined, versus just 5% of those who never hit one, more than four times the rate, per the same Greetwell survey. Handed a free afternoon in an unfamiliar city, respondents picked a friend's tip over the AI's by more than three to one, 31% to 9%. We'd trust the map search over the AI's confident tone, too.
What to do before you book anything AI suggested
Run every AI-suggested place through an independent check before it goes on your itinerary. We'd never book a deposit off a single AI citation, and 1 in 6 people getting sent to a fake place is exactly why.
- Search the exact name and city on Google Maps, not just the AI's own text
- Check the business's own site or a verified social account for today's hours
- Verify tours on the operator's own booking page, not just the AI's description
- Treat a suggestion with zero other mentions online as a red flag, not a bonus
- Keep anything unconfirmed flexible or refundable until you've checked it yourself
That takes minutes per stop, and it's what separates a usable itinerary from one that falls apart on day two. For the deeper argument on how far to let a chatbot go before booking, see our breakdown of what ChatGPT and Gemini actually do.
Screenshotting an AI itinerary and booking straight from it, with no independent check, is how a wrong-hours mistake turns into a wasted afternoon. Skip it.
This is honest error, not fraud, and we'd rather lose two minutes to a map search than an afternoon to a fake place. A cloned voicemail confirming a fake booking can take real money, a different problem we've covered separately.
Some planning tools sidestep part of this by design. TripProf's itinerary suggestions, for instance, pull from a structured database of real, addable places, so a suggested stop either exists there or it isn't offered. That's a different failure mode, not immunity: a smaller catalog can miss a real place, and it does nothing for the fully-booked or wrong-hours problems behind most of this survey's numbers.
How we got here
This survey is new, but the problem isn't. Documented cases go back to at least 2023, and they follow the same pattern: a confident answer, a real trip built around it, and a traveler finding out the hard way.
- March 2023 A Notion AI itinerary sent a Singaporean traveler to Ba Na Hills on day 8, actually near Da Nang, far from her Hanoi base.
- June 2023 The Points Guy's ChatGPT test recommended an unverifiable "magic show at Casino Lisboa" and sent a visit to Time Out Market on a day it was closed.
- July 2025 A Malaysian couple drove 370km (230 miles) to ride the "Kuak Skyride," a cable car that existed only in an AI-generated video, complete with a fabricated TV reporter.
A model built to sound helpful fills a gap with something plausible rather than saying "I don't know." That's not a reason to avoid AI for travel. It's a reason to treat its output as a draft, like a tip from a stranger on a forum. For the fuller case, see why AI can't actually plan a trip on its own.
What to watch next
No follow-up survey or regulatory response is tied to this report, and inventing one would be dishonest. Worth tracking instead: whether assistants start verifying against live maps before naming a business, and whether next year's survey shows the rate improving as the tools mature.
Frequently Asked Questions
Can you trust AI travel recommendations?
Not blindly. 55% of AI-using travelers hit at least one wrong recommendation in the same Greetwell survey, so treat any AI-suggested place as a lead to verify, not a confirmed booking.
Which AI tool is most likely to make up a fake destination?
The survey didn't break results out by tool. Documented cases span a chat assistant and an AI video generator, so this looks like a category-wide pattern, not one bad actor.
How do I know if an AI-recommended place is real before I book it?
Search the exact name on Google Maps and check the business's own site for current hours, rather than trusting the AI's description alone.
Is it safe to book a hotel or tour based on an AI suggestion?
It's fine as a starting point, not a final answer. In the same Greetwell survey, 64% said they have little or no trust booking a hotel through AI, and 63% said the same about a tour. Treat the AI's pick as a shortlist candidate, not a confirmed reservation.
What should I double-check before following an AI-built itinerary?
Hours and availability first, the two most common failure types in the survey, at 27% and 21%. Then check the stops are geographically realistic; AI itineraries have put destinations hundreds of kilometers apart on the same day.
Key takeaways
- 1 in 6 is the headline, not the whole story. 55% hit some kind of wrong recommendation, and wrong hours or prices are more common than invented places.
- Trust drops hard after one bad experience. Travelers who hit a bad recommendation were about four times more likely to say their trust in AI fell.
- Verify before you build a day around it. A map search and the venue's own site catch most of what this survey measured, in minutes per stop.
- This is an old problem with a new number attached. Documented cases of AI-invented destinations go back to at least 2023.
- Grounded tools help but don't replace the check. A planner like TripProf that suggests from a structured places database still can't fix a fully-booked venue, so the check still matters.
Sources
- Greetwell, "AI Travel Survey 2026" - primary survey, Aug. 2026
- KESQ / Stacker - Sept. 16, 2026
- WSB-TV Atlanta, same wire story
- WHIO Dayton, same wire story
- Deloitte, "2026 Travel Industry Outlook"
- AsiaOne, Notion AI itinerary case, March 2023
- NextShark, same case
- The Points Guy, ChatGPT test, June 2023
- PetaPixel, Kuak Skyride case, July 2025
- South China Morning Post, same case
- IBTimes UK, same case
- TechRadar, "AI is tricking people into traveling to places that don't exist"
Keep Reading
More travel tips and guides picked for you
Should You Let AI Book Your Trip? What ChatGPT and Gemini Actually Do in 2026
OpenAI pulled ChatGPT's travel checkout button, Google is testing AI-booked hotels, and one study caught chatbots pushing pricier sponsored flights. Here's the real 2026 verdict on letting AI handle the purchase.
Travel Blogging Myths 2026: 7 Debunked With Real Data
Most new travel bloggers earn $4.44 a month. AI search cut median blog traffic 85%. Here's what 2026 income surveys, traffic data, and real blogger accounts say about the myths still circulating.
AI Travel Scams 2026: Booking Sites, Voice Clones, and Fake Confirmations (And How to Stop Them)
AI travel scams 2026 are no longer obvious. The April Booking.com breach, FBI voice-clone warnings, and 900% surge in booking fraud changed the rules. Here's the 6-step verification checklist and family code-word system that actually stops them.