Would a Large Language Model Pay Extra for a View? Inferring Willingness to Pay from Subjective Choices

TL;DR

This paper investigates how large language models make subjective choices in travel scenarios, estimating their willingness to pay and comparing it to human benchmarks, revealing both potential and limitations of LLMs in decision support.

Contribution

It introduces a method to infer LLMs' willingness to pay from subjective choices and compares these estimates to human data, highlighting systematic deviations and factors affecting valuation accuracy.

Findings

Larger LLMs can produce meaningful WTP estimates.

Models tend to overestimate human WTP, especially with expensive options.

Conditioning on prior preferences improves valuation accuracy.

Abstract

As Large Language Models (LLMs) are increasingly deployed in applications such as travel assistance and purchasing support, they are often required to make subjective choices on behalf of users in settings where no objectively correct answer exists. We study LLM decision-making in a travel-assistant context by presenting models with choice dilemmas and analyzing their responses using multinomial logit models to derive implied willingness to pay (WTP) estimates. These WTP values are subsequently compared to human benchmark values from the economics literature. In addition to a baseline setting, we examine how model behavior changes under more realistic conditions, including the provision of information about users' past choices and persona-based prompting. Our results show that while meaningful WTP values can be derived for larger LLMs, they also display systematic deviations at the attribute level. Additionally, they tend to overestimate human WTP overall, particularly when expensive options or business-oriented personas are introduced. Conditioning models on prior preferences for cheaper options yields valuations that are closer to human benchmarks. Overall, our findings highlight both the potential and the limitations of using LLMs for subjective decision support and underscore the importance of careful model selection, prompt design, and user representation when deploying such systems in practice.