Comparative Accuracy of Generative AI in Bibliographic Cataloging Analysis According to MARC21 Standards and AACR2 Rules

Abstract

Purpose: This research aims to conduct a comparative study of the accuracy of Generative AI in analyzing bibliographic cataloging based on MARC21 standards and AACR2 rules.

Methods: This study employed an evaluation research design utilizing document analysis to describe the bibliographic characteristics of a sample of 30 books. The analysis focused on key bibliographic elements, including titles, authors, publication information, and International Standard Book Numbers (ISBNs). The data were analyzed using descriptive statistics, such as frequency and percentage.

Findings: ChatGPT demonstrated the highest overall performance, achieving an average accuracy score of 86.18%, followed by Claude (80.08%) and Gemini (73.17%). At the individual data-field level, ChatGPT achieved 100% accuracy across nine MARC21 fields: Tags 020, 050, 110, 245, 250, 260, 300, 650, and 700. Claude also showed high accuracy in basic and core MARC21 bibliographic fields, including Tags 020, 110, 245, 260, 300, 650, and 700; however, minor discrepancies were observed in certain fields requiring more complex contextual interpretation. In contrast, Gemini exhibited limitations in processing in certain subfields, with lower accuracy observed in fields such as Tags 020 and 050.

Applications of this study: Although Generative Artificial Intelligence (Generative AI) serves as a convenient tool to facilitate information resource cataloging, librarians or information professionals should carefully verify the accuracy of AI-generated bibliographic data before  practical application to maintain data quality, cataloging standards, and academic credibility.