How accurate is AI travel information?

A quarter of travellers have been given travel information by an AI that was wrong or out of date. Because the measure is self-reported, it captures errors that were caught; the true rate is higher.

The diagnosis matters for anyone trying to fix it. Most travel inaccuracy is not the model inventing facts from nothing. It is the model answering from stale or missing information: hours that changed, a property that renovated, a route that no longer runs. Where current structured facts are not reachable, a model answers from whatever it last saw, with the same confident tone.

This is why 42% always fact-check and only 6% fully trust. Travellers have calibrated to the error rate.

The practical response is not to argue with models. It is to make current facts machine-readable, structured and reachable, which is the discipline this observatory applies to itself, and the one it recommends.