Purpose This paper aims to present a case study of integrating generative artificial intelligence (AI) technologies into archival metadata creation in an academic library, providing an example of improving descriptive metadata for digitized photograph collections to enhance discoverability and accessibility. Design/methodology/approach This study tested various vision-enabled and multimodal large language models for generating Metadata Object Description Schema-compliant metadata from digitized photographic prints. Models were evaluated on visual interpretation and metadata output quality. Following a pilot test that validated this approach, Anthropic’s Claude Sonnet 4 model was selected for production implementation, creating descriptive metadata for 2,263 digitized photos. A Python-based process helped integrate this into existing workflows for digital collections. Findings The process provided substantial improvements over existing minimal descriptive metadata. Generated subject terms mapped successfully to the Faceted Application of Subject Terminology vocabulary in 64% of cases, demonstrating compatibility with established standards. Project team discussions led to the implementation of user transparency notices about AI involvement in metadata creation. Post-project analysis revealed this approach to be a cost-effective method of metadata enhancement. Practical implications This case study provides actionable guidance for cultural heritage institutions evaluating AI integration into their workflows and can be adapted for different institutional contexts and collection sizes. Originality/value While AI applications in libraries are expanding, this study documents a practical approach to integrating generative AI into archival metadata creation. The workflows, quality controls, cost analysis and transparency measures offer a roadmap for those considering adopting these technologies at scale.
J. Wood (Mon,) studied this question.