AI use in manuscript preparation may widen the open-access citation advantage

Citation counts play a major role in scholarly visibility and often influence academic evaluation and career advancement. Open access (OA) publishing has long increased this visibility by removing paywalls. As generative artificial intelligence (AI) becomes more integrated into literature discovery and manuscript preparation, this advantage may intensify because retrieval systems cannot reliably access full text behind paywalls. We analyzed 10,475 papers across publication sources and disciplines in which authors explicitly disclosed AI use; each was compared with two non-disclosed controls from the same source, similar fields, and nearby dates. We then measured the OA share of cited references. AI-disclosed papers cited a higher share of OA literature than matched controls (+2.49 percentage points; 95% CI, 2.08 to 2.92; P = 2.38 × 10−36), with the largest shift when AI was used for literature or research assistance (+5.98 percentage points; P = 4.76 × 10−9). The pattern held across dataset, impact-factor, and cited journal-year analyses. These findings indicate that machine accessibility is emerging as an additional visibility factor in reference selection. Because generative AI adoption in science remains at an early stage, this pressure may grow as retrieval and drafting systems become more widely used. The risk is especially relevant for underfunded researchers, institutions, and less-resourced fields, because high OA publication fees may make subscription or non-OA publication routes more likely and leave their work easier to miss in machine retrieval.