Asked to plan trips, AI chatbots sent travelers to places that do not exist. Among the 485 U.S. leisure travelers in the Greetwell AI Travel Survey 2026 who had used or tried AI for travel, 55% hit at least one recommendation that was inaccurate, unavailable, nonexistent, closed or otherwise wrong. Broken down: 27% got incorrect prices, hours or other details; 21% were pointed to something fully booked or unavailable; 17% got something simply wrong for what they asked; 16% were sent to a place, tour or activity that did not exist; and 10% were recommended a business that was permanently closed.
1 Answer
Expert: Greetwell (AI Travel Report 2026), Travel concierge company; August 2026 survey of 1,000 U.S. leisure travelers Greetwell's August 2026 survey of 1,000 U.S. leisure travelers - 485 of whom had used or tried AI for travel - found that these errors are not exotic edge cases. They are a routine output of AI trip planning, and they land exactly where a traveler has the least room to absorb them: an afternoon built around an activity that does not exist, a price or opening time that is wrong, a venue that is fully booked or has closed for good. The trust data show what that costs. Travelers who had hit at least one bad recommendation were more than four times as likely to report declining trust in AI travel recommendations as those who had not (21% against 5%). Trust and use also diverge at the point of commitment: 34% of respondents had used AI to research or plan a trip but not to book anything, against 14% who had used it for both. Majorities say they would have little or no trust in an AI tool choosing and booking without double-checking - flights 68%, rental cars 65%, hotels 64%, local tours or activities 63%, restaurant reservations 54% - and when asked to pick a single source for one free afternoon in an unfamiliar city, travelers chose friends or family (31%), hotel or local staff (25%) and online reviews (19%) well ahead of an AI assistant (9%). Read as a checklist rather than a headline, the failure list is specific: confirm that the venue or activity actually operates and is bookable on the date you want, that the price and opening hours are current, and that the operator is still taking bookings. A fluent, confident itinerary is not evidence that any of it exists. And the exposure is growing, not shrinking - Deloitte's 2026 travel research puts generative-AI use for trip planning at 25% of travelers, up from 15% a year earlier. Source: https://www.greetwell.com/us/research/ai-travel-survey-2026
Your answer
Sign in to verify this AI response.
Don't trust us — or the AI. Ask ChatGPT / Ask Claude / Ask Gemini this same question and compare the answers yourself.
More from this topic
Asked to cancel only the outbound leg of a round-trip SkyMiles award between New York JFK and Los Angeles while keeping the return, Delta Concierge told the passenger in writing: "Your JFK to LAX flight has been canceled, and your LAX to JFK flight remains booked." Delta's own refund confirmation shows the opposite: all 32,400 miles were redeposited and the full $11.20 in taxes refunded, with both DL752 (JFK-LAX) and DL938 (LAX-JFK) listed as canceled - the receipt even displayed "Refund Number: null". The AI confirmed an itinerary state that Delta's ticketing system had already destroyed, and Delta Reservations then told the passenger the original award could not be restored and that a replacement one-way would cost nearly twice as many miles.
A passenger asked United Airlines' customer-service chatbot how long she had to use a $200 United travel credit before it expired. The chatbot answered that the money would stay available for years. A screenshot of the exchange shows the bot stating that her $200 TravelBank credit "stays active for 5 years from the date it's deposited." The passenger, Alison Gil, planned around that answer and held the credit for a larger trip. A few weeks later United emailed her to warn that the same $200 in TravelBank cash would expire on September 27, 2026 - roughly one year after it was deposited, not five. When she questioned the discrepancy, she says United told her: "sorry, the bot told you that, but it's one year and we can't change it." The airline later acknowledged to NBC 5 Responds that its "AI chatbot provided this customer with the wrong information for her situation," while noting the disclaimer customers are shown saying information "may not always be complete or relevant."
ChatGPT was asked to plan a Victoria Falls trip covering Zimbabwe, Zambia and Botswana. It recommended the KAZA Univisa, which it said covered travel across all three countries with unlimited re-entry for 30 days - so the traveller built his entire multi-country itinerary around one visa and assumed he could cross borders freely for a month. In reality the KAZA Univisa is a joint Zambia-Zimbabwe visa: it becomes invalid if the holder leaves those two countries for one night or more, and it covers Botswana only for same-day trips through the Kazungula border.