A socio-technical exploration of bias in generative AI travel planners

The proliferation of generative AI has transformed how travelers plan and evaluate destinations, but it also raises urgent questions about potential bias. This study develops a socio-technical framework to examine biases in generative AI travel planners through three complementary studies. Study 1 synthesizes prior works to derive a taxonomy of seven bias categories spanning technical, informational, and social subsystems. Study 2 empirically audits 5000 AI-generated itineraries, measuring the prevalence, severity, and co-occurrence of these biases across models and prompt frames. Study 3 explores user perceptions through interviews, showing how biases emerge at the intersections of subsystem dynamics. Together, the findings demonstrate that biases in generative travel planning differ in form, mechanism, and consequence from those in traditional recommender systems and LLM research. Biases in generative AI travel planners are not isolated technical errors but emergent properties of socio-technical systems. Theoretically, the study advances socio-technical bias research by introducing a model-agnostic taxonomy and demonstrating cross-subsystem interactions. Practically, it highlights implications for tourism stakeholders, including the need for oversight of AI-based travel planners, safeguards against inequitable visibility of destinations, and careful governance of tourism data used in AI training.