Abstract
Institutional repositories are widely proposed as a source of African material for training artificial intelligence systems, on the assumption that what they hold is digital, open and retrievable. This paper audits that assumption for Nigerian university repositories. A census of the OpenDOAR directory was combined with direct retrieval of twelve repositories in August 2026, addressing registration, reachability, holdings, metadata quality and reuse rights. The census returned thirty-nine entries reducing to thirty-two universities, while three operating repositories were found outside the directory entirely, indicating undercounting. Five of nine sampled repositories responded to an automated agent. Holdings stood in thousands rather than millions, a quarter of academic units at one university held nothing, a fifth of records at another carried no date, and one abstract had lost all word spacing. No repository published a machine-readable licence, and one prohibited robotic harvesting. The paper concludes that the deposit gap is a sequence of compounding filters rather than a shortage of scholarship.
Keywords: Artificial Intelligence, Institutional Repositories, Nigeria, Open Access, Text and Data Mining
Publication Date: 2026-10-31