Unlucky Seven

A publisher reports that up to one submission in seven arriving on its desk is probably the product of a paper mill.

Not a predatory journal. A publisher. Real journals, real peer review, real indexation.

That figure is from a 2025 study in PNAS, and — believe it or not — it is the least alarming thing in it.

What the study found

Richardson and colleagues at Northwestern and the University of Sydney set out to map the networks behind industrial-scale research fraud.

Paper mills are usually described as factories producing fake papers. The study's finding is that they are better understood as brokerages — and that the brokering does not stop at manufacture.

They identified twenty-two editors at PLOS ONE who were anomalous for accepting articles that were later retracted, and thirty-three who were anomalous for articles later flagged on PubPeer.

They identified twenty-one authors whose papers were repeatedly assigned to the first group, and eighteen to the second.

And they found that paper mill agents have attempted to bribe journal editors.

Paper-mill output is doubling roughly every eighteen months. Scientific publication as a whole doubles about every fifteen years.

Retractions are rising, and not remotely fast enough to keep pace.

Why this is different from what you have been told

The standard advice about research fraud is: check the journal.

Look at the editorial board. Verify the identifiers. Watch for a near-identical title. We publish that advice ourselves, and it is good advice.

It assumes the journal is the unit of risk.

This study shows the compromise can be individual editors inside a journal that is entirely legitimate. PLOS ONE is real. Its peer review is real. Every check on our own journal-checking page would pass it, because there is nothing wrong with it.

You cannot detect a compromised editor by examining the journal.

How large the market is

A 2022–23 survey of medical residents at tertiary hospitals in southwest China found 46.7% self-reported buying or selling papers, letting other people write papers, or writing papers for others.

Nearly half.

That is not a fringe. It is a functioning market, and it exists because the pressure that creates demand is real. Papers lead to recognition, recognition leads to jobs, and the supply of jobs has not grown with the supply of researchers.

The study's authors make that point directly: unequal access to resources is part of what drives the behaviour.

What it means for your bibliography

A reference can be entirely real and still be worthless.

Real DOI. Real journal. Real named authors. Indexed where indexation counts. And built on data that was never collected.

Every check you can run from outside will pass it.

This is exceedingly uncomfortable, because there is no procedure that fixes it. The tools we all use — reference managers, DOI resolvers, journal checkers — verify that a thing exists and is what it says it is. None of them can tell you the research was done.

The advice you will be given, and why it fails

Ask many what protects you here and you will get this:

> Read the paper. Not the abstract, not the citation as it appeared in somebody else's reference list — the paper. A source you have actually read cannot surprise you later, because you know what it claims, how it got there, and whether the method could support the conclusion. A bibliography of forty sources you have read is worth more than a hundred you have collected.

It sounds like wisdom. It, or one of its variations, is given thoughtlessly and all too often — and it is the more dangerous for it.

A fabricated paper is internally coherent. That is what makes it saleable.

The method section describes a plausible method. The data look like data. The statistics are consistent with the data. The conclusions follow from the numbers. All of it hangs together, because it was written to hang together.

Reading it carefully confirms that it is coherent. It cannot tell you the experiment happened.

That is the same failure we described in a previous blog post, Seven of Nine. A check that examines a thing against itself tells you about internal consistency and nothing else — and reading a single paper is exactly that check, performed by a person instead of a machine.

You can read forty fabrications with great care and be no better off than somebody who collected a hundred.

What actually protects you

Nothing at the level of a single source. The protection is structural.

No claim your work depends on should rest on one paper.

A mill can manufacture a paper. It cannot manufacture a literature. Fabricated findings do not replicate, because there was nothing to replicate — so a claim supported by independent groups, using different methods, arriving from different directions, is expensive to fake in a way a single paper is not.

Where a finding stands alone and everything downstream depends on it, that is where to look hardest.

And look outward rather than inward.

PubPeer — has anybody raised concerns about this paper, these authors, this journal?

Retraction Watch — has anything from this author or group of authors been withdrawn?

Citations that engage rather than list. A paper cited forty times in passing tells you less than one cited twice by people who argued with it.

None of that is available from inside the paper, which is the point. The checks that work are the ones that compare the source to something outside itself.

And it is about to get harder

A model retrieving literature does not distinguish a paper-mill product from real research. It sees a real DOI in a real journal, indexed, cited.

So a candidate asking an AI for sources can be handed fabricated research with impeccable credentials — and the AI has no more ability to check it than a reference manager does.

The tools that scale are exactly the tools that cannot tell.

On checking a journal, when the journal is the problem: Is this journal real?

On a bibliography you no longer trust: Purging Citation Contamination.

Both free. No paywall, no freewall (gack... a freewall? Come on!). So? No signup, no email address, nothing to unlock.

Richardson, R.A.K., Hong, S.S., Byrne, J.A., Stoeger, T., & Amaral, L.A.N. (2025). The entities enabling scientific fraud at scale are large, resilient, and growing rapidly. PNAS, 122(32), e2420092122. Full text, open access.