Using ChatGPT with prompt engineering for personalized travel destination recommendations

Abstract

Artificial intelligence (AI) has become widely accessible and can support a wide range of use cases such as personalized recommendations. Our study investigates the capabilities of a pre-trained large language model for generating personalized travel destination recommendations. Based on a design science research approach, we evaluate the usability and effectiveness of suggestions. In particular, we explore how additional task-specific or user-specific input can enhance the response effectiveness. User preferences gathered from a survey are used for computational experiments. The results are presented in a second survey, and the participants’ feedback is collected to assess the perceived quality of personalized recommendations. This feedback demonstrates that the model recommendations are positively evaluated with regard to user satisfaction, preference consideration, and timesaving. Yet, the participants also see potential for more detailed and specific recommendations, along with enhanced granularity for the budget breakdown and better transparency through shared sources. We also identify limitations, such as the availability and quality of gathered data, leading to potentially inadequate recommendations, as well as subjective evaluation measures requiring larger and more diverse sample data to confirm the generalizability of the approach. The findings indicate the potential benefits of personalized AI recommendations in the travel industry and suggest areas for improvement.