Making Hidden Collections Visible: A Case Study of Human–AI Collaboration in Archival Metadata Generation

Abstract

Archives and special collections increasingly face an imbalance between the rapid expansion of digitized
materials and the slower pace of descriptive metadata creation needed to support discovery and access.
This article presents a case study of a pilot project conducted at American University Library to evaluate
JSTOR Seeklight, an artificial intelligence (AI)-assisted tool designed to generate descriptive metadata
and transcription for digitized archival materials. Academic libraries have long served as stewards of
unique and irreplaceable primary resources housed in archives and special collections. A substantial
portion of these materials remains confined to physical formats, discoverable only through limited or
incomplete description and accessible primarily to on-site researchers.1 Despite decades of professional
attention to improving access and discovery of these materials, progress has often been incremental.
At the same time, the scope of digital stewardship has expanded well beyond the initial act of resource
digitization.