Horse Diagnosis and Triage Accuracy of GPT-4o

Abstract

Animal owners may increasingly rely on large language models for gathering animal health information alongside internet sources in the future. This study therefore aims to provide initial results on the accuracy of ChatGPT-4o in triage and tentative diagnostics, using horses as a case study. Ten test vignettes were used to prompt situation assessments from the tool, which were then compared to original assessments made by a veterinary specialist for horses. The most probable diagnosis suggested by ChatGPT-4o was found to be quite accurate in most cases, with the urgency to contact a veterinarian sometimes assessed as higher than necessary. When provided with all relevant information, the tool does not seem to compromise horse health by recommending excessively long waiting times, although there is still potential for improving the relief of veterinarians' workload.