One Wrong Answer: What the Research Says About Trust and Intent

It is easy to assume that travellers treat AI answers with a pinch of salt — that a wrong price or a closed facility quoted by an assistant is shrugged off as the machine’s problem. The best evidence available says otherwise.

The study

Kim, Kim, Park, Kim, Jhang and King, publishing in the Journal of Travel Research, tested how travellers respond when a generative AI assistant gives them inaccurate information during tourism decision-making.¹ Participants received AI-generated answers about a destination; some answers contained an inaccuracy, and the researchers measured the effect on two things that matter commercially: intention to visit, and trust.

Both fell after a single inaccurate answer. The effect was strongest when the error was prominent in the answer — the kind of detail a traveller would notice and act on — rather than buried. In short, one wrong fact was enough to move behaviour, and the more visible the fact, the larger the move.

Why this changes the priority order

Most organisations, when they first read what assistants say about them, want to fix the most egregious error. The research suggests a different rule: fix the errors travellers are most likely to rely on, because reliance is where trust and intent are lost.

That reorders the list. A price is prominent and relied upon; independent testing has found every major assistant understating hotel prices,² which means a great many travellers are arriving at a booking page expecting to pay less than they will. Inclusions, cancellation terms, opening hours and access are relied upon at the moment of decision. A slightly dated description of the décor is not.

The traveller blames you

The second finding of practical importance is attribution. Participants did not separate the assistant’s error from the organisation it described. The disappointment attached to the destination. This matches what front-line teams already report: the guest who was told breakfast was included does not complain to the assistant.

Combined with what we know about where assistants get their facts — overwhelmingly from third-party review platforms, intermediaries and listings rather than the organisation itself³ — the picture is uncomfortable. An organisation can be held responsible for a statement it never made, sourced from a page it does not control, and only find out when the traveller arrives.

What to measure

The study gives organisations a defensible reason to treat answer accuracy as a standing measure rather than a one-off audit. A practical version: collect the answers assistants give for the ten questions travellers ask most about you, mark each fact accurate, inaccurate, incomplete or out of date, rank the inaccuracies by how likely a traveller is to act on them, and correct at source in that order. Repeat on a cycle, because answers drift.

It is the same discipline any organisation applies to a published price list. The difference is that the price list is now being read aloud to customers by a system you do not control — and, as this research shows, they are listening.