Do travelers trust AI differently for brainstorming versus visa tasks?

McKinsey and Skift's survey of 1,002 travelers finds that trust in AI-based travel tools is high in aggregate — "more than 90 percent of customers report some confidence in the accuracy of travel information that they receive through AI" — but that this trust is not uniform across task types. The report states that "consumers tend to express more comfort with using currently available AI tools to brainstorm in open-ended ways (such as finding inspiration or destination ideas) than with using these tools in more high-stakes situations (such as understanding visa requirements and resolving customer service issues)."

McKinsey and Skift tie this gap to the gen-AI-era reputation for hallucination: the report notes the wariness toward high-stakes AI use is "understandable, especially given the pervasive reports of hallucinations occurring in gen AI output." The distinction maps onto the report's broader framing of gen AI as an "adviser" offering "well-informed counsel" versus agentic AI as a "direct report" that can "autonomously act" on decisions — travelers appear willing to accept advisory-grade uncertainty for low-stakes brainstorming but not for consequential, fact-bound tasks where an autonomous agent might act on a wrong answer.

This complicates the framing in Does the personal assistant model actually serve most users?, which argues that broad personal assistant for everyone framings over-generalize from a narrow segment wanting automation of routine tasks; the McKinsey/Skift finding suggests the relevant split is not only which users want automation but which task types, within the same traveler, earn enough trust to be handed to an agent. It also bears on Why do AI agents fail at workplace social interaction?: both findings locate agent limits in task character rather than in raw model capability, though one measures stated trust and the other measures completed task performance.

The finding is self-reported comfort from a survey, not a measured error rate or an observed outcome when travelers actually delegated a high-stakes task to an agent, so it does not show that AI agents are in fact less reliable on visa or customer-service questions, only that travelers perceive them that way. The report does not quantify how large the comfort gap is between the two task categories. The implication it supports is that trust in travel AI is task-calibrated rather than uniform, which bears on where in the booking or service journey agentic AI can be deployed without first addressing, or disclosing, hallucination risk.