Dramatic increases in redundant publications in the Generative AI era

Area of Science:

• Scientific integrity
• Bibliometrics
• Health research ethics

Background:

• Redundant publication distorts the scientific record and wastes resources.
• Increased use of large open-science data, like NHANES, raises concerns about AI-facilitated redundant manuscripts.
• Generative AI (GenAI) may be enabling the creation of formulaic, redundant publications.

Purpose of the Study:

• To quantify redundant publications from a single large health dataset (NHANES).
• To investigate if GenAI can produce synthetic manuscripts that evade standard integrity checks.
• To assess the effectiveness of current plagiarism detection in the GenAI era.

Main Methods:

• Systematic search for redundant publications using NHANES data (2021-2025).
• Definition of redundancy: same exposures, outcomes, and national population.
• GenAI prompted to create synthetic manuscripts designed to evade detection; tested with plagiarism software.

Main Results:

• Identified 411 redundant publications across 156 exposure-outcome pairings.
• Example: NHANES data on oxidative balance score and chronic kidney disease published 6 times in 1 year.
• GenAI-generated manuscripts showed low similarity scores (14-26%), evading typical plagiarism thresholds.

Conclusions:

• A 17-fold increase in redundant publications (2022-2024) signals a systemic failure in editorial checks.
• Redundant papers distort research (meta-analyses, scientometrics) and waste peer review resources.
• Current plagiarism and redundancy checks are inadequate for GenAI; publishers need enhanced cooperation and guidelines.