NASIG 2026

Libraries have long grappled with the complexities of legacy metadata, particularly in serials and government documents, where inconsistent practices, evolving cataloging standards, and repeated system migrations create persistent data quality problems. Tasks once viewed as too time-consuming or difficult to tackle at scale are becoming increasingly manageable with the rise of practical, library-focused applications of artificial intelligence. This session presents two case studies from the University of Toronto Libraries that demonstrate how AI can transform long-standing metadata challenges into transparent, replicable, and sustainable workflows. The first case study highlights an initiative that uses ChatGPT and Alma Cloud Apps to generate and standardize serials item-level metadata. Through AI-driven methods, the project automates the retrospective itemization of holdings, creating more accurate and consistent records. It also demonstrates how generative AI can support both the creation of item-level metadata and the regeneration of serial summary holdings. The second case study illustrates how AI-generated Python scripts can remediate government document metadata at scale by clustering similar bibliographic records and programmatically selecting higher-quality OCLC replacements to enhance discovery and overall record quality. Together, these projects show how generative AI, guided by librarian oversight and supported by robust APIs, can amplify human expertise and make long-standing legacy data challenges newly solvable.