Finding · Confirmed fabrications
0
citations in the biomedical literature that point to studies which do not exist. Verified across PubMed, Crossref, OpenAlex, Google Scholar.
CaseCIT-2026-01
Status● Open
WindowJan 2023 — Feb 2026
CorpusPubMed Central OA · 2.5M papers

References
to nowhere.

An automated audit of citation integrity across 2.5 million biomedical papers uncovers a rising signature of references that cite studies which do not exist — spreading through the peer-reviewed literature at more than twelve times the rate of three years ago.

Topaz · Roguin · Gupta · Zhang · Peltonen
Columbia University Irving Medical Center · The Lancet, Vol 407, May 9, 2026
Scroll
Growth
>12×
Fabricated citations per 10,000 papers, 2023 → early 2026.
Prevalence
1 : 458
Papers containing at least one fabricated citation, by 2025.
Corpus
2.5M
Biomedical papers audited · 125.6M references extracted.
Most affected single paper
18 / 30
References fabricated in one published study (Ren et al., 2025).
§ 01 / The TrendFig. 1

A more than 12× rise in three years. No precedent.

The quarterly rate per 10,000 papers indexed in PubMed Central rose from approximately four in 2023 to 56.9 in early 2026, a more than twelvefold increase, with no precedent in the recorded history of citation error.

~4/10k
Baseline — 2023
56.9/10k
Latest — early 2026
>1,200%
Change over audit window
Fabricated citations · quarterly rate
PMC OA · 2023 → early 2026 · n = 2.5M
§ 02 / The ScaleFig. 2

From 125 million references to 4,046 fabrications.

Every reference in the corpus was verified against four independent databases. The funnel below traces how 125.6M references narrowed to 4,046 confirmed fabrications.

Step 01References extracted
125,600,000100.000 %
Step 02With verifiable identifiers
97,100,00077.310 %
Step 03Flagged for mismatch
30,8120.024 %
Step 04 · ConfirmedFabricated citations
4,0460.0032 %
§ 03 / The AnatomyInteractive

Can you spot the fake?

Fabricated citations are engineered to look indistinguishable from real ones: real-sounding authors, plausible journals, credible dates. The pair below are real citations from our audit — one a verified, indexed publication, the other a fabrication identified in the published Lancet supplementary appendix. Select the one you believe is fabricated, then reveal the verdict.

Prompt — Click the fabricated citation
Citation A
Verified
Walters WH, Wilder EI
"Fabrication and errors in the bibliographic citations generated by ChatGPT."
Sci Rep. 2023;13:14045. PMID 37679503.
Claimed title found in PubMed (PMID 37679503)
Title matches the record at that PMID
Authors and journal verified across PubMed, Crossref, OpenAlex
DOI 10.1038/s41598-023-41032-5 resolves to the same paper
Citation B
Fabricated
Chen F, Liu Y, Wang H, Zhang X, Li J, Yang K, et al.
"Microglial Modulation via Cannabinoid Receptor 2 Alleviates Fibromyalgia-Related Pain."
Neuroscience. 2023;519:14–28. PMID 36813155.
Claimed title returns 0 results in PubMed, Crossref, OpenAlex, and Google Scholar
PMID 36813155 resolves to a different paper: "ChatGPT in Research: Balancing Ethics, Transparency and Advancement" (Graf & Bernardi, Neuroscience, 2023)
DOI resolves to the same unrelated ChatGPT-ethics editorial
Claimed authors have no joint publication on cannabinoid receptors
Published as Example C in the Lancet supplementary appendix

Both citations read as equally professional. Only automated verification across FOUR INDEPENDENT DATABASES can reliably tell them apart. This fabrication pattern — plausible title attached to a real PMID that resolves to an unrelated paper in the same journal and year — is the dominant signature in our dataset.

§ 04 / The Case StudyRen et al., 2025

One paper. Thirty references. Eighteen fabricated.

A 2025 paper on ureteroileal anastomotic techniques in an open access oncology journal contained 18 (60%) fabricated references of 30 verified. Each fabricated reference was tailored to the paper's narrow surgical topic, attributed to real urologists, and bore claimed publication years of 2023 or 2024. The claimed titles do not correspond to any publication indexed in PubMed, Google Scholar, or CrossRef — yet each carries a real PMID and DOI that resolves to a different, unrelated paper. Concentration at this level is the upper tail of what we observed across 2.5 million biomedical papers in this audit.

Ren C, Xiao M, Zhu J, Tong W, Yi F. Front. Oncol. 15, 1613772 (2025)
Fabricated · 18 Not flagged · 12

Citation pattern suggests systematic rather than incidental fabrication. Source paper ↗

§ 05 / The ConcentrationFig. 3

Not all publishers are equal.

Fabricated citations are not uniformly distributed. The highest-rate publisher produced fabrications at more than fourteen times the rate of the most selective journals in the dataset.

Rate per 10,000 papers · top publishers (anonymized)
Publisher A
56.8
Publisher B
40.9
Publisher C
21.6
Publisher D
13.7
Publisher E
8.0
Publisher F
5.1
Most selective
3.9
Ratio · Highest vs. most selective
14×

More than a third of all fabricated citations came from just two publishers.

The concentration is not random. It tracks editorial intensity, peer-review depth, and the economics of fast-turnaround publishing.

§ 06 / Human-ReadableIn plain terms

What the data means.

1/458

By 2025, one in every 458 biomedical papers contained at least one fabricated citation.

18/30

The most affected single paper had eighteen of thirty references pointing to studies that do not exist.

≥ 3

Many affected papers contain three or more fabricated citations — evidence of systematic, not incidental, fabrication.

Estimated in real time
0

fabricated citations have entered the biomedical literature since you opened this page.

Session 00:00 Rate · 56.9 / 10,000 Flux · ~1,600 papers/day

The literature does not correct itself.