Paper Mills in 2026: What the BuyTheBy Dataset Reveals About Research Fraud

On April 23, 2026, Retraction Watch covered a preprint by Reese Richardson at Northwestern University that had been circulating in the research integrity community for several days. The dataset, called BuyTheBy, compiles 18,710 text-based advertisements from seven paper mills operating in India, Iraq, Uzbekistan, Latvia, Ukraine, Russia, and Kazakhstan. Each advertisement is an offer to sell authorship on a scientific paper, at prices ranging from roughly $57 to over $5,600 per slot. The median price for a first authorship slot is close to US $800. The dataset covers advertisements collected between March 2020 and April 2026 and includes 51,812 time-stamped price data points across more than 5,500 unique products in 14 different categories.

The BuyTheBy preprint does not simply describe a problem researchers have long suspected. It documents that problem systematically, with transaction records, price histories, and product types that demonstrate a mature, organized commercial market. Nature's journalists reviewed more than 600 advertisements linked to around 400 articles and identified 53 published papers with titles matching those in the dataset. Of those 53, only five had been retracted as of the time of reporting. The majority are still in the peer-reviewed record.

For researchers submitting honest work, the paper mill problem is not an abstract ethics concern. It affects the reliability of the literature you cite, the journals where your own manuscripts compete for review slots, and the downstream credibility of fields you contribute to. This article covers what is currently known about how paper mills operate, what the recent evidence says about scale, and what practical steps are available to researchers who want to protect their own work from contamination.

From Cottage Industry to Industrial Operation

Paper mills have existed in some form since at least the early 2000s, but their profile shifted sharply in the 2010s when open access publishing lowered the marginal cost of a citable article and when publication counts became a primary metric for academic advancement in several national systems. A mill in that earlier era typically sold a small number of ghostwritten papers targeting minor journals with weak peer review. The economics were relatively simple: a researcher at an institution where promotion depended on publication output would pay for papers that the institution could not produce internally through legitimate work.

What makes the current situation different is scale and operational sophistication. The BuyTheBy dataset covers seven mills, and these are not all the active operations. Each one documented in the dataset functions more like a publishing services company than a freelance fraud operation: they advertise specific journals, specific authorship positions, and specific turnaround times. The product catalog includes first authorship slots, last authorship slots, middle authorship positions, and in some cases spots on review articles where a purchased author's name appears alongside legitimate contributors who may not know the arrangement exists. Some advertisements correspond to papers already accepted for publication. Others are pre-production, offering a slot on a manuscript that will be submitted after the buyer pays.

The geographic distribution of the mills in BuyTheBy reflects market adaptation. Operations in South Asian markets tend to advertise at lower price points and target journals in engineering, computer science, and basic biology. Operations in Eastern Europe and Central Asia have historically concentrated on medical journals, where publication value in career terms is higher and prices reflect that. The dataset captures, for the first time, that these are not isolated vendors operating independently. There are patterns of pricing, geographic specialization, and product diversification that look more like a supply chain than a collection of individual fraudsters.

The BMJ Study: Nearly 10 Percent of Cancer Papers Flagged

In January 2026, The BMJ published a study that applied a machine learning model to the entire body of cancer research indexed between 1999 and 2024. The model was trained on 2,202 papers already identified as paper mill products in the Retraction Watch database and used a fine-tuned BERT architecture to classify papers based on their titles and abstracts. Applied to 2,647,471 cancer studies, it flagged 261,245 papers as potentially originating from paper mills. That figure represents 9.87 percent of the cancer literature examined.

The model achieved a classification accuracy of 0.91 on its holdout test set. That is a useful result, but not a perfect one, and the authors were careful to describe it as a screening tool rather than a verdict. A flagged paper is not a confirmed paper mill product. What the model identifies is statistical similarity to known paper mill output, and any individual flag should be treated as a prompt for human investigation rather than a finding of misconduct.

With that caveat stated, the scale of the flags is striking. More than 170,000 of the flagged papers came from authors affiliated with Chinese institutions, accounting for 36 percent of China's cancer research output in the dataset. The BMJ study does not argue that this reflects a unique failure of any particular national system. It reflects where market pressure has been concentrated, where authorship-counting mechanisms created strong incentives to purchase, and where the paper mills that have been most thoroughly documented were primarily operating. Mills adapt to wherever institutional demand for publications is highest, and there is no reason to assume the pressure is limited to one country or one field.

Why High-Impact Journals Are Not Safe Either

One comfortable assumption researchers sometimes make is that paper mill contamination is limited to low-prestige journals with weak peer review. The evidence consistently challenges this. The BMJ study found that flagged papers appeared in the top 10 percent of journals by impact factor, not only in mid-tier venues. Nature's review of BuyTheBy advertisements found published matches in journals owned by Springer Nature and Wiley, as well as in IEEE conference proceedings. These are not obscure outlets.

This matters because it changes the practical risk assessment for researchers relying on the literature. If paper mill products were concentrated in journals with impact factors below 2 or in publishers outside the major commercial houses, the problem would be relatively contained. Instead, the evidence suggests that paper mills actively target journals with strong reputations precisely because placement there increases the product's value to the buyer. A fabricated paper in a recognized journal sells for a premium. An authorship slot in a high-impact venue is worth more on a CV than a slot in a minor title, and the price data in BuyTheBy reflects that.

Guest-edited special issues have been identified as a recurring vulnerability across multiple publishers. When a journal outsources editorial decision-making to an external guest editor, the normal gatekeeping processes become thinner. Several high-profile retractions in recent years have involved clusters of papers from a single special issue, suggesting that a compromised guest editor may have facilitated the placement of multiple mill products in a single batch. A BMJ Group journal retracted seven of eight papers from one guest-edited special issue in April 2026. Peer review integrity is harder to maintain when editorial oversight is distributed to parties who may not fully understand or respect that responsibility.

How AI Changed the Detection Problem

Paper mills were already difficult to detect before large language models were widely available. They typically involved real scientists, or people presenting as real scientists, submitting papers that had been substantially fabricated but that looked superficially credible to peer reviewers. Images were copied from other papers or digitally manipulated. Methods sections contained impossible experimental parameters. References were sometimes real but irrelevant, serving a citation-padding function.

Generative AI tools have made several parts of that process cheaper and harder to catch. A mill can now produce plausible scientific text at scale without employing large numbers of writers. AI image generation tools create figures that are harder to reverse-search than directly copied images. Plagiarism checkers that look for verbatim copying are less effective against text generated fresh for each submission. The Retraction Watch database, which now contains more than 63,000 retractions, shows a steady increase in AI-related integrity concerns, though the database does not cleanly separate AI-generated papers from other categories of misconduct.

There is an uncomfortable symmetry in this situation. The same AI tools that researchers might use legitimately to draft a methods section or clean up a translation are being used to industrialize fraud. That is not an argument against responsible AI use by honest researchers. It is an argument for why the disclosure and verification practices now required by most major publishers make sense: a documented, reviewed, human-accountable use of AI is categorically different from an undisclosed AI-generated submission, even if the two look similar on the surface. The ICMJE's January 2026 updated recommendations, which introduced a dedicated section on AI in publishing, reflect exactly this concern.

What the BuyTheBy Advertisements Reveal About the Market

Before BuyTheBy, researchers knew paper mills existed because they had been caught. The Retraction Watch database documented their output retroactively. What was harder to reconstruct was how the market actually functioned: what was sold, to whom, at what price, and with what targeting of specific journals. The dataset compiled by Richardson and colleagues fills part of that gap.

The 14 product categories in the dataset include first authorship, last authorship, middle authorship, citation insertion (paying to have a reference added to a paper you did not author), review article slots, and in a smaller number of cases, entirely ghostwritten papers delivered to the buyer ready for submission. Prices vary by product and geography. A middle authorship slot in a lower-tier journal was available for as little as $57 in some markets. A first authorship slot targeting a high-profile venue cost upward of $5,600. Citation insertion, where a broker arranges for a specific paper to be cited in an upcoming publication, was available at lower price points than full authorship, which helps explain how citation counts can be artificially inflated without a corresponding spike in full authorship fraud.

Richardson's stated intent is for BuyTheBy to be a starting point for journals, publishers, and other authorities to take action, rather than a comprehensive final analysis in itself. Several of the advertisement titles in the dataset correspond to papers that remain in the peer-reviewed record. Of the 53 published papers identified by Nature's journalists in their review of a sample of roughly 600 advertisements, only five had been retracted as of the time of reporting. The gap between advertisement volume and retraction rate suggests that a large proportion of paper mill output is never removed after publication.

Practical Steps for Researchers

If you work in a field with documented paper mill activity, which at this point includes most of the biomedical sciences and a growing proportion of engineering and computer science, there are concrete practices worth building into your research workflow.

The most direct protection is checking papers before you cite them, especially papers from authors or institutions that appear repeatedly in suspicious contexts. The Retraction Watch database is publicly accessible and searchable. PubPeer, an online platform for post-publication review, often carries integrity concerns about papers long before any formal editorial action is taken. The Problematic Paper Screener is a free tool that cross-references papers against known retraction and flag databases and returns a summary of concerns. None of these tools can guarantee that an unchecked paper is clean, but they catch a meaningful proportion of papers with documented problems.

When you find a retracted or flagged paper already in your reference list, update the citation before submission. This is not merely a courtesy. Journals increasingly check submitted reference lists against retraction databases as part of routine desk review. Citing a retracted paper without noting the retraction signals to reviewers and editors that you have not kept up with the literature your work depends on. In some journals, it is now flagged as a compliance issue that stalls processing.

Pay particular attention to review articles in your reference list. Paper mills often target authorship slots on review papers because a single review article can generate many downstream citations through its reference list. If a review article claims to synthesize a large body of evidence and appears in a journal without strong editorial oversight, it is worth checking whether any of the primary papers it cites have integrity flags before you rely on its synthesis. Review articles propagate errors efficiently, and paper mill products mixed into a review's citations travel into every paper that subsequently cites the review.

If you are preparing a systematic review or meta-analysis in a field with known paper mill exposure, integrity screening should now be part of your eligibility assessment, not an afterthought. A meta-analysis that aggregates paper mill products alongside genuine studies is not just imprecise. It is building on a contaminated foundation, and the results carry that contamination into any clinical or policy decision that relies on them. This is most acute in areas like cancer biology, hepatology, and cardiology, where the BMJ study and prior research have identified the highest concentration of suspected fraudulent output.

What Publishers and Journals Are Doing

Publishers have been aware of paper mills long enough that several have invested significantly in detection infrastructure. Elsevier, Springer Nature, Wiley, and the major society publishers have all described internal screening programs, though the specifics are rarely disclosed in detail. COPE, the Committee on Publication Ethics, has published guidance on paper mill investigations and maintains retraction guidelines that describe expected procedures when a cluster of suspicious papers appears in a single journal. Those guidelines require timely investigation, editorial transparency with readers, and coordination with institutional research integrity offices.

Detection tools in use by publishers and integrity researchers include reverse image search to identify reused or manipulated figures, reference list analysis to find citation pattern anomalies, authorship network analysis to spot unusual co-authorship clusters, and increasingly the kind of ML screening applied in the BMJ cancer study. Several publisher groups have established dedicated integrity teams whose job is to investigate flagged papers and coordinate retractions. The Retraction Watch database, now at over 63,000 entries, provides the historical training data that makes these tools possible.

None of these measures is sufficient on its own, and the publishers know it. The consensus from integrity researchers is that the volume of paper mill output far exceeds what current detection catches. One widely cited estimate holds that the vast majority of paper mill products have not been detected. That gap between what is in the record and what has been identified has real consequences for every researcher who uses that record as a foundation.

The BuyTheBy dataset gives journals and publishers something they did not previously have: a documented, cross-referenced list of specific products offered for sale, some of which correspond to papers that can now be actively investigated. Richardson's framing of the dataset as a starting point for authorities is deliberate. The intent is for publishers to use the advertisement records to initiate investigations into specific papers, rather than waiting for integrity sleuths to do that work reactively and piecemeal.

The Implication for Your Own Submissions

Honest researchers sometimes worry that widespread fraud makes their own work harder to evaluate fairly. The more a journal editor has been exposed to paper mill submissions, the more cautious they may become about papers arriving from certain countries, certain institutions, or certain research areas. That concern is legitimate, and individual researchers cannot fully address it through their own actions alone.

What you can address is how your manuscript presents itself. Mills produce papers quickly, with thin methods, suspicious figures, generic results discussions, and author lists that have no traceable academic history outside the paper at hand. A manuscript that provides detailed, reproducible methods; figures with clear provenance; a reference list that has been checked for integrity; acknowledged limitations; and authors with real, findable academic profiles looks materially different from what comes out of the operations documented in BuyTheBy. Those differences are perceptible to careful editors and reviewers, even without automated detection tools.

The BuyTheBy dataset, the BMJ cancer study, and the Nature investigation together represent the most detailed picture yet of the paper mill market. They do not overstate what is known. A flagged paper is not a proven fraud. An advertisement is not a confirmed sale. But taken together, they indicate that the scale of the problem is large, that it reaches into journals most researchers would consider reputable, and that current detection and retraction systems are catching only a fraction of what is present. For a working researcher, that is not a reason for despair. It is a reason to be systematic about what you cite, transparent about how you work, and deliberate about the journals you submit to.

Further Reading