AI hallucinations in tourism: How errors impact consumer trust and recommendation acceptance?

Abstract

Generative artificial intelligence (GenAI) is quickly transforming travel planning; however, its outputs can include hallucina-tions, which are plausible yet false statements that can undermine user judgement. Eliminating hallucinations in GenAI tech-nology is currently impossible. This research analyses, through two experiments, how hallucinations in itineraries generatedby ChatGPT influence consumer behaviour. Study 1 (n = 1004) explores a serial-mediation pathway where the presence of hal-lucinations affects the intention to follow the itinerary through perceived accuracy, perceived usefulness and trustworthiness.Hallucinations significantly reduced perceived accuracy. As accuracy increased, usefulness improved, leading to greater trust-worthiness, which strongly predicted intention. Although the direct effect of hallucinations on intention was not significant,the serial indirect effect was negative and significant. Study 2 (n = 241) investigates how the outcome's positive or negative valueand the importance of error feedback influence the results. The findings indicate that the valence of outcomes moderates therelationship (i.e., positive outcomes yielded higher intention to follow the recommendation than negative outcomes). However,when errors were salient (i.e., when there was a hallucination accompanied by negative consequences) intentions to follow therecommendation decreased further. Together, these studies contribute to an integrative framework that connects cognitive fac-tors (accuracy, usefulness), attitudinal factors (trust), experiential factors (observed outcomes) and contextual factors (error sa-lience) to explain how and when GenAI hallucinations can reduce compliance with travel advice. The findings enhance ourunderstanding of the potential drawbacks of GenAI in tourism and provide actionable guidance for developing more transparent,reliable and user-centred GenAI travel systems.