In a national online survey commissioned by travel‑concierge firm Greetwell, 1,000 U.S. adults who took at least one overnight leisure trip in the past year were asked about their use of artificial‑intelligence tools for travel planning.
AI adoption is growing, but accuracy lags
Overall, 53% of respondents reported using or trying AI for travel. Deloitte’s 2026 travel research confirms the trend, noting AI usage rose from 15% to 25% within a year. Yet the survey revealed a significant flaw: 1 in 6 AI‑using travelers (approximately 17%) were recommended a destination, tour or activity that does not exist.
More than half of AI users (55%) reported encountering at least one bad recommendation of any kind. The most common failure types included incorrect addresses and events that never occurred, wasting travelers’ time and planning effort.
Trust in AI remains mixed
When asked about overall trust, 34% said their confidence in AI had increased over the past year, while 14% said it had declined. However, among those who experienced a problematic recommendation, 21% reported a drop in trust compared with only 5% of respondents who had not faced any issues.
The survey also showed that most travelers are reluctant to let AI handle bookings without supervision. Only 14% used AI to both research and book a trip, while 34% used it solely for research and 5% tried it but found it unhelpful.
Human guidance still preferred
When presented with a scenario of a free afternoon in an unfamiliar city, 31% said they would rely on friends or family for suggestions, versus just 9% who would turn to AI. Additionally, 65% indicated they would want a real person to contact for rebooking or refunds if weather forced a cancellation.
Greetwell’s findings suggest that while AI can speed up travel planning, many consumers still value the human touch for reliable recommendations and support. The company notes that the survey cannot prove a causal link between AI use and trust levels, but it raises questions about how far travelers will let AI operate before demanding verification.
The study was conducted in August 2026, with 485 respondents forming the base for the accuracy and trust‑gap analysis. Percentages were weighted at the respondent level and may not sum to 100% due to rounding.
Original reporting: El Paso News (HLL/CB) — read the source article.
