Paper Mills: How Editors and Institutions Detect Them

Paper mill manuscripts are built to pass the checks journals rely on. Catching them takes editors who read patterns and institutions that verify what only they can see.

Explainer · 10 min read · Directive Publications Editorial Team

What a paper mill is and what it sells

A paper mill is a business that sells publication. The COPE and STM research report on paper mills describes the activity as "the process by which manufactured manuscripts are submitted to a journal for a fee on behalf of researchers with the purpose of providing an easy publication for them, or to offer authorship for sale." COPE's flowchart defines the wider behaviour rather than the business: systematic manipulation is where individuals have repeatedly used dishonest or fraudulent practices, with the goal of influencing the publication record and/or achieving financial gain.

The product is not always a wholly invented paper. A mill may supply:

• a finished manuscript with fabricated or plagiarised content
• an author position on a paper already moving through review
• favourable peer review, written through reviewer accounts it controls
• placement in a guest-edited special issue it has taken over

Mills can also pass as author services, their websites opening with editing or translation before they reach guaranteed publication. That gives you the test to teach: a legitimate service improves work you did; a mill supplies the work, the authorship or the review.

The stakes in clinical medicine are not reputational. Fabricated findings get built on by other researchers, absorb funding, and can reach patients.

How a fabricated manuscript reaches acceptance

This is the process publishers most commonly see:

1. Mill staff write a paper and submit it to several journals at once.
2. When one journal accepts it in principle or asks for revisions, the other submissions are dropped.
3. The mill advertises up to six author places, priced by the journal's impact factor and the position in the byline.
4. Buyers are generally given an email address, and the mill handles the correspondence.
5. In some cases the mill suggests reviewers who then return positive reports.
6. Once a journal publishes one paper, many similar submissions can follow.

COPE's guidance on systematic manipulation of the publication process explains the reviewer step: the suggested reviewers may be real experts or invented names, but the email addresses are falsified, so invitations go back to the manipulators, who write the reports.

Guest-edited collections attract mills for a reason the COPE and STM report sets out: a special guest-edited issue is a popular route to publication, and that creates an opening, as special issues and guest editing explains. Keep reviewer-identity verification with the publisher, not the guest editor who proposed the collection.

The finished manuscript generally follows the author guidelines closely, is written in good English and passes plagiarism checks. A clean similarity result is not clearance, which is why reading a similarity report properly is only one layer of screening. Mills also move away from journals that systematically reject their papers, so consistent rejection deters repeat targeting.

The warning signs editors look for

The indicators below combine the infographics in COPE's systematic manipulation guidance with the publisher list in the COPE and STM report. Read them as a set, not a scorecard: most have an innocent explanation alone, and mean something only when the same thing shows up in a second manuscript.

In the submission and authorship record

• the same manuscript submitted more than once, or many submissions from one group in a short period
• author emails that do not match the author's name, several accounts linked to one address, or different authors submitting from the same IP address
• authors with no publishing record and newly created ORCID iDs
• requests to change the author list after acceptance, especially the corresponding author

In the manuscript itself

• unrelated papers that share wording, layout, figure fonts and error-bar styles
• Western blots that look "too clean", without molecular weight markers, or image elements that also appear in another paper
• results that do not match the stated aim, or reagents that cannot do what the methods claim
• missing ethical approval for animal experiments
• affiliations that name no specific university, or departments unrelated to the topic

In peer review

• the same reviewers suggested by different authors
• reviewer addresses that are non-institutional and cannot be verified
• reports returned unusually fast
• reports for different manuscripts that share formatting and phrasing

One item needs the caveat the report itself supplies. It calls non-institutional email "a particularly challenging indicator as there are countries where institutional email addresses are rare and therefore that does not automatically mean the author is not genuine." Weight it by geography, never alone.

An indicator you can only see once is one you cannot use: matching IP addresses, repeated reviewer suggestions and shared figure styling are invisible one file at a time, so these checks have to run against your submission history. COPE's flowchart also grades recommended actions by the editor's confidence; read the current version directly.

Tortured phrases and other text fingerprints

Guillaume Cabanac, Cyril Labbé and Alexander Magazinov coined the term in a 2021 study of tortured phrases, describing them as "unexpected weird phrases in lieu of established ones." An established term is replaced word by word with synonyms: "counterfeit consciousness" for artificial intelligence. The authors found free online tools that produce exactly this output, and some articles they examined contained other papers' abstracts rewritten this way.

What a tortured phrase evidences is automated rewriting, a different finding from a purchased paper. Two rules keep the judgement fair:

• Separate it from non-native English. Awkward grammar is common and innocent; a standard term nobody in the field would phrase that way is different, though a few flagged phrases can be legitimate in context.
• When you find one, look for more. One odd phrase is a query for the authors; several, alongside the indicators above, is a pattern worth comparing against likely source texts.

Cabanac and colleagues also run the Problematic Paper Screener, which scans published literature for tortured phrases and other signs of generated text so suspect papers can be reassessed.

Biomedical papers have their own fingerprints. A 2019 study describing the Seek & Blastn tool reported that incorrectly identified nucleotide sequence reagents were a frequent feature of a cohort of highly similar gene knockdown papers. A reviewer in a molecular field can check whether the sequences in the methods match the targets claimed.

What happens once a journal suspects a paper mill

Under COPE's flowcharts, before publication the editor suspends peer review, gathers evidence and asks the authors to explain. Where data look suspicious the editor can request raw files such as uncropped blot images, having decided in advance how they will be judged and by whom. If the response is missing or unconvincing, the journal asks the authors' institutions to investigate. If an institution stays silent, COPE suggests following up every three months and, with no response a year after first contact, rejecting the affected manuscripts.

After publication, the threshold has moved. COPE's retraction guidelines, revised in September 2025, treat "any form of misrepresentation" as grounds for retraction and name paper mill involvement explicitly, alongside identity theft and fictitious authorship. They also say retraction should not be delayed because "the authors or their institutions are not cooperative or responding promptly", so the journal's timetable is now largely independent of yours. Where concerns are substantial and credible but there is not yet enough to decide between correction and retraction, an expression of concern alerts readers while the investigation runs.

Because mills work at scale, COPE's 2023 supplemental guidance also lets editors handle linked articles as a batch. If a journal writes to you about a set of papers, expect it to differ from a single-article query:

• Decisions can rest on patterns across articles, including circumstantial evidence, without a "smoking gun".
• Data requests are at the editor's discretion, because genuine and fabricated files are hard to tell apart without help from whoever holds the originals.
• Authors may be asked whether any third party helped with the research, writing, submission or revision.
• The journal can proceed without institutional input.

The journal decides whether to correct the published record on publication-process evidence alone. Establishing who did what — which author knew, who paid, whether anyone was added without their knowledge — is a finding only an employer can make.

What institutions can check that journals cannot

A journal sees a manuscript and its metadata. You hold employment records, ethics committee files, lab notebooks and access to the authors, so these checks are easy for you and impossible for an editor.

• Ethics approvals. Confirm that the approval reference cited in a paper exists in your committee's records and covers that study.
• Affiliations. Check that the stated department and role match the author's actual position, and look for papers claiming your institution for people you do not employ.
• Collaborations. One mill website quoted in the report told buyers its articles were the collective work of authors from different universities and could not include grant or funding details. Unrelated co-authors with no shared project deserve a question.
• Raw data. Ask the author for instrument files, uncropped images and dated records. Someone who did the work can produce them.
• Timing. A burst of papers outside someone's usual field before a promotion deadline is a reason to ask, not evidence.

When a journal writes to you, answer. Give a named contact, confirm receipt, preserve the relevant data and correspondence, and tell the editor what you can and cannot establish. The research integrity office checklist covers intake and preservation step by step.

Detection is the smaller half of the job. The report's first recommendation is aimed at institutions and funders rather than journals: change the incentives that push researchers towards mills, such as doctoral students who cannot graduate without a publication. If your progression rules demand output without supervision or protected time, you are creating the demand a mill exists to serve.

What to tell your researchers

Put three plain rules into induction and supervisor training:

1. An offer of authorship or guaranteed publication for a fee is a paper mill offer, however professional the website looks. Decline it and report it to the research integrity office.
2. If your name appears on a paper you did not work on, or an author list changes without your agreement, tell the journal and the research integrity office in writing.
3. Keep the records that show the work is yours: raw files, dated notebooks, ethics approvals and contribution statements.

Add that buying a place does not buy a way out: the COPE and STM report notes that the warranties such authors sign are void, but the responsibility stays with them.

Where detection actually fails

Paper mills survive on the assumption that somebody else has already looked. The journal assumes an employer vets its staff; the employer assumes peer review catches fabrication; the supervisor assumes the co-authors did the parts they did not see.

So the useful question for a research office is what it does with a journal's letter today. If those land in a shared inbox and take a month to reach someone who can pull an ethics file, the retraction will be decided without your evidence.

You can read how Directive Publications describes its peer review process and its guidance for authors.