Fake Citations In Science Papers Nearly Quadrupled Since 2023

Fake Citations Are Rising in Science, and Few Get Fixed

Science runs on receipts. When a researcher makes a claim, a citation points to the earlier study that backs it up. Lately, more of those citations lead nowhere. One in every 435 peer-reviewed papers published in early 2026 in physics, chemistry, materials science, engineering, and computer science included at least one fabricated reference, up from one in 1,600 in 2023. These references point to studies that researchers couldn’t locate in major academic databases or through online searches.

That finding comes from a scan of 500,000 journal papers from January 2023 through March 2026, all vetted by outside experts before publication. Fake references had been documented mainly in medical journals. This analysis, reported in the Proceedings of the National Academy of Sciences (PNAS) and built on OpenAlex, an open database of scholarly work, found them in every field it checked. The rate climbed steadily, reaching about one in 650 in 2025.

Almost none of the affected papers had been retracted or corrected when researchers checked. Paper mills, operations that churn out sham research, can produce fake references, as can other misconduct and careless use of AI writing tools. AI tools can generate references that look real but aren’t.

Automated Checks Found Fake Citations in Hundreds of Papers

Researchers drew their papers at random, separately from each of the five fields, and extracted 26.2 million references. They kept the roughly 82% that carried a digital identifier, a barcode-style code linking a reference to a specific record online. The detection system was adapted from the team’s earlier audit of biomedical papers.

Software compared the title each reference listed with the title actually tied to its identifier. Mismatches went through more filters, including an AI language model that separated simple formatting slips from genuinely made-up citations. A title that couldn’t be found in any of seven sources, including Google Scholar and a general web search, was labeled fabricated. That label describes whether something can be verified, not whether an author meant to deceive anyone. In all, 1,135 references across 619 papers got the label.

To test accuracy, two independent experts reviewed 400 flagged references without knowing how the software had classified them. They confirmed 85.8% as fabricated, so roughly one in seven flagged references may not be truly fake.

Physics Has the Highest Rate of Fake Citations, Chemistry the Lowest

Across the full study period, about 26 of every 10,000 physics papers contained at least one fabricated reference, compared with about 4 of every 10,000 chemistry papers. Materials science ranked second, followed by computer science and engineering. Chemistry’s figure is likely an undercount because some chemistry journals are filed under a different category in the database.

Scaling those rates up to the roughly 7 million papers indexed in these fields over the same period suggests about 8,500 papers in the database carry at least one fabricated reference. Most had one or two. Of the 619 affected papers, 518 had exactly one. But 27 papers had five or more, and that small group accounted for more than a third of all the fabricated references found.

Researchers offer several possible reasons for the gap between fields, though none has been proven. Fields with faster publishing pipelines would show the problem sooner within the study window. Physics also leans heavily on arXiv, a free site where researchers post papers before outside review, so an unfindable reference may look less suspicious.

Nearly All Papers With Fake Citations Still Stand Uncorrected

Researchers checked public records for retractions among the affected papers. Only two of the 619 had been retracted, meaning formally pulled from the published record, and neither was among the 27 papers with the most fabricated references. One retraction notice cited fabricated references alongside other miscited sources. The other involved concerns about authorship and incomplete data.

That leaves 617 papers with no retraction, correction, or warning at the time of the audit. Researchers caution that this is a snapshot, since many of the papers are recent and fixes can take time. They also note that one or two fabricated references may be a simple citation error, so each paper needs a case-by-case judgment from its publisher. The numbers hint at a problem that may be growing faster than the system meant to catch it.

Researchers did not blame AI writing tools for the rise. The timing fits the spread of those tools, but months-long delays between submission and publication, along with other factors, make it impossible to pin the increase on a single cause.

Automated Reference Checks Could Catch Fake Citations at Submission

A citation is a promise that evidence exists somewhere, ready for anyone to check. A fake one breaks that promise and chips away at the trust science depends on. The authors recommend automated reference checks when papers are submitted, plus screening of papers already published, and want fabricated references tracked as their own category, separate from plagiarism and data falsification. A fabricated reference can be caught by checking it against records of published work, which leaves publishers little reason to let them slip through.

---

Disclaimer: This article reports on findings from a single published research paper and is intended for general information only. The results reflect the authors’ methods and samples and may change as further research emerges.

---