Evaluating Data Availability Statements in Academic Literature: A Scalable AI Approach

The growing emphasis on research transparency and data availability, driven by open science principles and FAIR data standards, underscores the need for systematic monitoring of data-sharing practices. This study develops an AI-powered framework to detect and evaluate Data Availability Statements (DAS) across 7301 journal articles from the fields of medicine and engineering (2015–2024). Leveraging PDF parsing, heuristic section detection, and large language model (LLM) classification with DeepSeek-V3, our automated method achieves high accuracy (99.6%) and F1 scores (> 0.97). Analysis reveals a significant rise in DAS prevalence post-2020, though most studies omit direct links to repositories. Medical articles frequently cite ethical restrictions to justify “available upon request” clauses, while engineering articles more often include persistent identifiers or GitHub links. Publisher policies also influence the inclusion of DAS in articles, with PLOS outperforming Elsevier in DAS completeness in the sampled dataset. By providing a scalable tool for auditing data-sharing practices, this study advances open science efforts, offering actionable insights to strengthen open science practice.