Scientists Keep Citing Papers That Don't Exist. You Can Guess Why

Title: "Fabricated citations: an audit across 2·5 million biomedical papers"

Journal: The Lancet (Volume 407, Issue 10541, pages 1779-1781)

Authors: Topaz M, Roguin N, Gupta P, Zhang Z, Peltonen L-M

Papers scanned: 2,471,758, 125.6 million total references extracted

Fabricated references: 4,046 across 2,810 papers

Time period: January 1, 2023 to February 18, 2026

Published: May 9, 2026

The study analyzed 97.1 million references across 2.5 million biomedical papers and found that fabricated references have increased more than 12x since 2023, with 1 in 277 papers in early 2026 containing dead-end citations.

With numbers like that, the ChatGPT-doctor memes may become a reality sooner than we think.

The Method

The researchers built an AI system that flagged differences between a citation's title and the actual title of the paper it pointed to. If a reference's title didn't appear in PubMed, Crossref, OpenAlex, or Google Scholar, it was classified as fabricated. But why just these four?

They cover the entire academic publishing landscape in a way no single database could, because each one has blind spots:

• PubMed only does biomedicine so a reference to a statistics paper in a medical journal won't be there.
• Crossref only covers works with DOIs. Older papers and many preprints (early drafts) don't have DOIs.
• OpenAlex is good but still building coverage for non-English content.
• Google Scholar is the most comprehensive, but it's less precise and includes scam journals.

Three humans manually checked 500 suspicious references and 7 out of 10 were confirmed fakes. The automated system was a bit better, with 91% accuracy overall. The numbers are bleak, and the authors say they're just scratching the surface.

The Usual Suspect

The spike started around mid-2024, which is the same time people started using LLMs to write and edit papers. Coincidence? Probably not. Earlier studies found that 30-69% of references generated by AI in biomedical papers are completely made up even though they look real. Correct formatting, real author names, believable dates, but peer review misses them.

Which Papers Are Most Affected?

Review articles. They are 57% more likely to have fake references than other types.

An oncology paper from 2025 had 18 out of 30 references that led nowhere. The fake citations were on a very specific surgical topic, credited to real urologists, with believable publication years. So someone put work into this.

The study also found 28 clinical trials and 79 systematic reviews with fabricated references. Topaz (lead author) says some of those fake citations are being cited by other papers and showing up in systematic reviews that doctors rely on to make decisions.

Retraction Debate

Even the experts can't agree on what to do.

The old-school editors say retract the whole paper if there's a hallucinated reference anywhere in it.

The researchers say it depends on whether the fake reference actually matters to the paper's conclusions. Most papers with this problem have just one or two bad citations, and those are probably honest mistakes from people who used AI without checking its work. So, correction and transparency, not retraction.

As of February 2026, 98% of the affected papers had seen no action from publishers. Only 1.6% had been retracted or corrected. None of the retractions explicitly mentioned fake references.

Should You Care?

Yes. Science is not the kind of field where these results should be a thing.

One fabricated reference could get cited by another paper, then another, then end up in a systematic review that shapes how patients are treated. Medical professionals have no way of knowing the evidence they rely on doesn't exist.

The authors mention that AI is used in the research process too, so depending on your views there's a black pot and a black kettle somewhere in all of this. Links below for the science people.

The Solution

Are you a scientist? Check your references before submitting. If you're a peer reviewer, spot-check a few random citations. There are free citation checkers that can help you speed through this process. Here are a few you can use, or replace with options that fit your workflow:

• sciwrite-lint: Checks references against CrossRef, OpenAlex, Semantic Scholar, and checks for retractions too. Also checks if cited papers actually support your claims. Runs entirely on your own machine so no manuscripts leave your computer.
• Lazy Scholar: This is a free Chrome extension that automatically checks citations as you browse academic articles. It shows retraction alerts, scam warnings, and also checks references against CrossRef.
• VeriExCite: This one's a free web app. Upload a PDF and it extracts the bibliography and checks each reference against Crossref, Google Scholar, arXiv, and Google Search. Shows which citations are verified, invalid, or not found.

This is worth doing properly.