Evaluating U.S. Federal Zoonotic Records Against National One Health Goals: A Multi-LLM Assessment Across COVID-19, Monkeypox, and H5N1

(1) Background: Zoonotic diseases require coordinated action across human, animal, and environmental health sectors. While substantial research has focused on outbreak prediction, forecasting, and surveillance, less attention has been given to systematically evaluating how One Health principles are reflected in government policy documents. (2) Methods: We developed a multi-large language model (LLM) framework to evaluate 477 U.S. government documents covering COVID-19, monkeypox, and highly pathogenic avian influenza (H5N1). Agency technical guidance and Federal Register documents were evaluated using a 19-item rubric covering three goals of the U.S. National One Health Framework: coordination and communication, outbreak response, and surveillance. Claude, GPT, and Gemini independently evaluated the documents. Inter-model agreement and human validation were assessed, and sensitivity analyses examined alternative chunk-to-document aggregation thresholds and the potential influence of document length. (3) Results: One Health alignment was generally limited, with no disease × document-type × goal combination exceeding 39%. Document type showed the clearest differences, with agency technical guidance showing approximately 13–31 percentage-point higher alignment than Federal Register documents across the three goals, while differences across diseases were less consistent. Sensitivity analyses supported the main document-type pattern across alternative aggregation thresholds and after restricting Type B documents by length. After excluding one wording-sensitive rubric item, the three LLMs reached unanimous agreement in 87.1% of document-item evaluations, with human–model agreement (Cohen’s κ = 0.513–0.692) comparable to human–human agreement (κ = 0.649). (4) Conclusions: This study introduces a purpose-built policy corpus and structured multi-LLM evaluation framework for examining how One Health principles are reflected in government documents. The clearest differences were observed across document types rather than diseases, and the main findings were generally stable across alternative analytical assumptions. Together, these findings extend zoonotic-disease research beyond outbreak prediction and surveillance toward systematic evaluation of One Health policy alignment.