Journalology #148: Fabricated citations
Hello fellow journalologists,
Earlier this week I sent paid subscribers a detailed breakdown of all the announcements from April. If you missed that overview, you can read it here.
Today’s newsletter is in the classic Journalology format. I’ve selected seven stories from the past two weeks that I think you should know about:
5% of articles have at least one fabricated reference
The merger of two not-for-profits
The revised FT50 journal rankings in economics and business
Wiley’s new leader of the Research group
A Forensic Scientometrics report
Elsevier’s lawsuit against Meta
Five years of Cell Press Multi-Journal Submission
Fabricated citations: an audit across 2·5 million biomedical papers
Among 97·1 million verified references, we identified 4046 fabricated references across 2810 papers. In 2023, approximately one in 2828 papers contained at least one fabricated reference. By 2025, this had risen to one in 458 and in the first 7 weeks of 2026, one in 277 papers had at least one fabricated reference. The fabrication rate increased more than 12 times, from approximately four per 10 000 papers in 2023, to 51·3 per 10 000 papers in the fourth quarter of 2025, reaching 56·9 per 10 000 papers in early 2026.
JB: This research article (which was published as a Correspondence letter in The Lancet; more on that later) provides further evidence that LLMs are introducing errors into the academic literature. Around 5% of articles have at least one fabricated reference, the authors found. The analysis was based on PubMedCentral data, so the results may not be generalisable outside of the biomedical sciences.
The authors propose that publishers should integrate automated reference verification into editorial workflows and indeed some do just that. This is just one integrity check, among many, that publishers now need to perform, which adds to the cost and complexity of academic publishing at a time when some funders are implementing APC price caps.
It’s possible to buy off-the-shelf solutions — or to get access to some of the tools via the STM Integrity Hub — but we shouldn’t underestimate the size of the research integrity challenge. It seems unlikely that volunteer-run Diamond OA journals, which some believe are the solution to the crisis in communication, will be able to keep up.
The large commercial publishers have invested heavily in new editorial technology in recent years. For example, as I outlined in a recent annual report deep dive, Springer Nature spent €188 million on technology between 2021 and 2025, and now has around 60 AI ‘assists’ supporting editorial workflows (N.B. not all of the €188 million investment will be into journals workflows). This week, IOP Publishing, the largest physics society publisher, announced that it has developed a tool to detect duplicate peer review reports.
Smaller publishers will struggle to implement these multiple research integrity checks. This could mean that fabricated content will be increasingly published by lower volume publishers, which don’t have the bandwidth to detect fabricated references, manipulated images, paper mills, reviewer mills etc. etc.
We should also consider the implications for preprint servers. Are we OK with a ‘rough and ready’ low cost approach for preprints? Should they be expected to invest in research integrity solutions too?
The Lancet authors recommend four actions:
First, publishers should integrate automated reference verification into submission workflows before peer review begins; verification tools exist, and the barrier to adoption is institutional rather than technological.
Second, indexing services should add integrity metadata to article records so that downstream users can assess the reliability of references.
Third, publishers should retroactively screen existing publications and issue corrections or retractions when fabricated references compromise a paper’s conclusions.
Fourth, fabricated references do not currently exist as a discrete category in major research integrity databases; establishing this category would enable systematic tracking and accountability.
The underlying assumption here is that publishers and indexers have sole responsibility for sorting out this mess.
A fifth recommendation should be added. Academic institutions bear some responsibility if their faculty produces AI slop. Angela Cochran, who leads the publishing team at ASCO, wrote an essay in 2024 that made exactly this point. Unfortunately, the institutions that most need to implement these technologies are the ones least likely to do so.
Let’s be clear what the root cause of the problem is here: a publish or perish academic culture where researchers feel they need to use AI tools, which can introduce errors, in order to be competitive. Researchers are under huge pressure from the institutions that employ them to be productive; they cut corners because there’s little comeback for doing so.
Finally, it’s worth considering why this piece of research was published as a letter to the editor, with a 9 page PDF appendix (!), rather than as a full research article. Presumably The Lancet’s editors didn’t want it to be a citable item in the impact factor calculations. However, as Ella Flemyng, the head of editorial policy and research integrity at Cochrane, notes in Retraction Watch’s story about this paper:
Though the approach [using AI] was validated on 500 records and the main limitations are discussed, we are lacking considerable details about the methods.
The impact of AI on scholarly communication is wide ranging. We need a coordinated approach. If your organisation hasn’t done so already, you can submit feedback on the STM Association’s consultation document Toward Responsible Use of Research Content in Generative Al until the middle of June.
