SGCAT: Using AI to Facilitate Cataloguing

Abstract

Cataloguing is traditionally a manual act of metadata creation that takes substantial focused manhours. Conventionally, libraries have relied on several approaches to increase efficiency of metadata creation, such as sharing of records via z39.50 protocol or via cooperatives like OCLC. However, cataloguing an item from scratch is still needed when they are new to library databases. Librarians must wade through stacks of new items individually to create valuable metadata according to comprehensive standards (e.g. MARC21, RDA) so that resources can be discovered by users. This is effortful and time-consuming, often taking anywhere from 30 minutes to hours depending on the item’s complexity. We developed a custom GPT prototype (“SGCAT”) to streamline bibliographic metadata creation and enhance efficiency in delivering library materials to patrons. Powered by OpenAI GPT, SGCAT is customised to follow specific cataloguing rules and local library practices, effectively serving as a smart cataloguing assistant. To ground it in fact, SGCAT pulls relevant bibliographic data from trusted sources such as NLB’s vendor-provided order information, Google Books and Open Library APIs. Through rigorous prompt engineering, SGCAT is currently able to draft a MARC record from a single ISBN in seconds, potentially speeding up cataloguing by at least 2x in combination with human review. SGCAT can follow instructions, maintaining consistency with cataloguing syntax and standards. By automatically generating informative abstracts, SGCAT takes over the workload from time-consuming transcription and summarisation tasks. These assistive capabilities help transform the cataloguer’s role from creator to reviewer, freeing them to focus more on cerebral tasks like subject cataloguing and record refinement. SGCAT has potential to be an assistant that elevates metadata quality and efficiency, speeding up time-to-shelf and improving discovery for patrons. The team intends to explore extending this prototype by enriching its knowledge base with more API sources, resolving , and enhancing SGCAT with multilanguage capability and multimodal input features to process cover images in the input prompts.

(presented on 15 August 2025 at "Metadata's New Frontiers: AI-Driven Systems and Standards" session)