---
title: Nonexistent Citations Are Entering the Scientific Literature Faster Than Journals Catch Them
description: An audit of 111 million references estimates 146,932 nonexistent citations in 2025 alone, and journal rules put the responsibility on authors.
author: Darie Nani (Editor-in-Chief)
updated: 2026-08-02T19:19:54.628Z
canonical: https://www.sovereignmagazine.com/article/nonexistent-citations-scientific-literature-2025-audit
image: https://cdn.nanimediahouse.com/hallucinated-citations-scientific-literature-88725.webp
categories: Science & Tech
content_type: News
region: Global
publication: Sovereign Magazine
schema_type: Article
---

[An audit of 111 million references](https://arxiv.org/abs/2605.07723) has found that citations pointing to papers that were never written are now embedded across the scientific literature, with a conservative estimate of 146,932 of them in 2025 alone.

Zhenyue Zhao, Yihe Wang, Toby Stuart, Mathijs De Vaan, Paul Ginsparg and Yian Yin drew their sample from arXiv, bioRxiv, SSRN and PubMed Central, covering 2.5 million papers, and report a sharp rise in references to nonexistent work following the widespread adoption of large language models. They chose citations for the audit because a reference is, in their words, a uniquely verifiable object: it either matches a real paper or it does not.

## The Errors Cluster Where AI Use Is Heaviest

The nonexistent references are diffusely embedded across many papers rather than concentrated in a few, but three settings stand out. Errors are more pronounced in research fields with rapid AI uptake, in manuscripts carrying linguistic signatures of AI-assisted writing, and among small and early-career author teams.

## The Credit Goes to Scholars Who Already Have It

Hallucinated references disproportionately assign credit to scholars who are already prominent, and to men. The authors say the pattern may reinforce existing inequities in scientific recognition.

## Preprint and Journal Checks Catch Only a Fraction

Preprint moderation and journal publication processes capture only a fraction of these errors, the authors report, and the spread of hallucinated content has outpaced the safeguards already in place. Their own paper is a preprint, posted on 8 May 2026, and has not been peer reviewed.

On 2 August, WisPaper, an academic research platform operated by Atom Infinite of Singapore, announced a citation checker it calls True Cite, which the company says lets researchers check references against real academic publications.

## The Rules Put the Responsibility on the Author, Not the Tool

The International Committee of Medical Journal Editors, which sets recommendations for medical journals, says journals [should require authors to disclose at submission](https://www.icmje.org/recommendations/browse/artificial-intelligence/ai-use-by-authors.html) whether they used AI-assisted technologies, and that chatbots and other AI-assisted tools should not be listed as authors, because they cannot be responsible for the accuracy, integrity and originality of the work.

On references its guidance is direct: "Humans must ensure there is appropriate attribution of all quoted material, including full citations." Authors are also expected to review and edit AI-generated content themselves, because the output can be incorrect, incomplete or biased. Nondisclosure of AI use may require corrective action, and in some circumstances may be construed as misconduct.

## Editors Are Told to Ask Whether AI Was Used

The recommendations also tell editors to query authors about AI use, and to be vigilant that authors may not disclose it when asked. They draw a line between uses: spell-checking and grammar correction are generally viewed as acceptable, while using AI to generate content has to be considered in the context of the work.

> "Referencing AI-generated material as the primary source is not acceptable."
> — ICMJE Recommendations, Use of AI by Authors

## FAQ

**Q: Can an AI tool be listed as an author on a research paper?**
No. The ICMJE recommendations say chatbots and other AI-assisted tools should not be listed as authors, because they cannot be responsible for the accuracy, integrity and originality of the work, and those responsibilities are required for authorship.

**Q: What do journals require authors to disclose about AI use?**
The ICMJE says journals should require authors to disclose at submission whether they used AI-assisted technologies, and to describe how they used them. It also says nondisclosure may require corrective action and may be construed as misconduct in some circumstances.

**Q: Has the citation audit been peer reviewed?**
No. It was posted as a preprint on 8 May 2026 and has not been through peer review.

**Q: Is using AI to check spelling and grammar treated the same as using it to write?**
Not quite. The ICMJE says using AI to spell-check and correct grammar is generally viewed as acceptable, while using AI for content generation has to be considered in the context of the work.