Enhancing Digital Library Accessibility through Artificial Intelligence for Individuals with Disabilities: A Systematic Literature Review

Abstract

Digital libraries are now the main route to scholarly content, yet their content, interfaces and services still exclude many disabled users, and promotion of artificial intelligence (AI) as a remedy has outpaced evidence that it works for them. Framed by the social model of disability and the Convention on the Rights of Persons with Disabilities (CRPD), this systematic literature review, reported according to PRISMA 2020, examined which AI technologies have been applied to digital library accessibility, what has been measured, and what follows for disabled users. Boolean searches of OpenAlex, Semantic Scholar, the Directory of Open-Access Journals, PubMed and ERIC identified 1,056 records; after removal of 429 duplicates and two-stage screening against explicit criteria (peer-reviewed, 2016–2025, English), 26 studies were included and synthesised thematically. Speech synthesis, speech recognition, natural language processing, document AI, optical character recognition, computer vision alt text and machine learning personalisation were studied, mostly for blind and print-disabled users. Coding every claim into three registers showed that 11 studies only describe what a technology does, 13 measure system or provider-side performance, and only two report outcomes measured with disabled users, both through self-report; this absence of outcome evidence is a principal finding. Audited conformance of scholarly documents is poor; disabled people appear mainly as beneficiaries, not co-designers. The review contributes an evidence-graded, five-stage integration model that specifies actors, gates and failure handling, and recommends treating AI outputs as unverified until evaluated with disabled users. Findings rest on single-reviewer screening and published records, and the model awaits empirical testing.