Large Language Models Yield Unsustainable Tourist Flows

This study quantifies the impact of generative AI on tourist flow using scenario-based projection models. We simulate one million US domestic tourists using Gemini 2.5 Flash and GPT 4.1 Nano. Their tourism patterns are compared to the null model and empirical-based simulations. Large language models generate tourist flows that are more seasonal, more unequal, and less reciprocal than empirical-based simulations. Geographic patterns, like mean travel distance and preference for neighboring states and intra-state destinations, vary by model. Findings show that the widespread adoption of generative AI can undermine the sustainability and resilience of tourism systems. We urge tourism scholars and practitioners to proactively assess the consequences of adopting generative AI in tourism.