•  
  •  
 

Abstract

This study evaluates the effectiveness of consumer-level AI tools—ChatGPT Pro and Microsoft Copilot—for generating descriptive metadata in digital archival collections. Using a pilot project at the University at Buffalo's University Archives, researchers tested these tools on transcripts from the WBFO radio archives to create program descriptions of varying lengths. Findings demonstrate that both tools significantly reduced processing time for audio collections, with each exhibiting distinct advantages: ChatGPT Pro offered more adaptable outputs while Microsoft Copilot excelled in structured environments with superior privacy protections. The study explores prompt engineering strategies, examines limitations including AI hallucinations when identifying subject headings, and addresses privacy considerations in institutional contexts. This research provides a scalable methodology for implementing AI in archival workflows while maintaining necessary oversight and quality control, contributing to broader discussions on AI integration in digital archiving practices.

DSC04896.jpg (1803 kB)
WBFO Reel-to-Reel Magnetic Media

DSC04855.jpg (4778 kB)
WBFO Program Guides to Recorded Audio

Share

COinS