AI-assisted taxonomic publishing: a human-supervised workflow and engineered prompt for Zootaxa authors

The accelerating crisis of global biodiversity loss demands a commensurate increase in the efficiency, accuracy, and rigor of taxonomic publishing. For over a decade, the recommendations presented by Dubois et al. (2011) have served as an important reference for authors publishing in Zootaxa, establishing a baseline for compliance with the International Code of Zoological Nomenclature (the Code). The recent emergence of Artificial Intelligence (AI), particularly Large Language Models (LLMs), presents a new opportunity to automate the verification of these complex nomenclatural standards. However, the integration of AI into this highly specialized field is fraught with significant risks, ranging from the hallucination of non-existent citations to the inability to parse undigitized historical literature. This paper presents a framework for AI-assisted taxonomic publishing, explicitly bifurcating the technology's utility into two distinct domains: (1) structural, syntactic, and internal consistency compliance, where AI demonstrates proficiency as a "compliance assistant," and (2) semantic, historical, and geospatial interpretation, where AI exhibits limitations that require human expertise. These boundaries are delineated, citing recent evidence regarding AI fabrication, geospatial reasoning deficits, and the structural barriers of the "taxonomic impediment". A "human-supervised hybrid workflow" is proposed that leverages non-AI database queries for comprehensive search tasks while exploiting LLMs for text analysis and formatting. Finally, an engineered prompt designed to function as a robust, pre-submission validation tool is provided, mitigating the risk of introducing unavailable nomina while acknowledging the irreplaceable role of the expert taxonomist.