In a Lancet Correspondence published this week, researchers reported finding 4,046 fabricated references across 2,810 papers after screening nearly 2.5 million open-access biomedical articles indexed in PubMed Central between January 2023 and February 2026.
The authors defined fabricated references as citations whose claimed titles could not be matched to any publication in PubMed, Crossref, OpenAlex, or Google Scholar after multiple verification steps.
The rate rose sharply during the study period. According to the audit, papers containing at least one fabricated reference increased from roughly 1 in 2,828 papers in 2023 to 1 in 458 papers in 2025, reaching approximately 1 in 277 papers during the first 7 weeks of 2026.
The quarterly fabrication rate increased more than twelvefold over the study window — from about 4 fabricated references per 10,000 papers in 2023 to nearly 57 per 10,000 papers in early 2026.
One example highlighted by the authors involved a 2025 oncology paper in which 18 of 30 verified references could not be matched to any real publication. The references appeared highly plausible: they were tailored to a narrow surgical topic, attributed to real urologists, and assigned believable publication years.
Why the Problem Is Difficult to Detect
The concern is not simply that fabricated references exist, but that many resemble legitimate citations closely enough to evade routine scrutiny.
The authors note that large language models are known to generate convincing but fictitious references, citing prior studies estimating that 30% to 69% of large language model -generated biomedical references may be fabricated. However, they caution that their audit was not designed to determine causation. Increased paper mill activity and changes in journal indexing practices may also contribute to the trend.
Most affected papers contained only one or two fabricated references, which the authors suggest may make detection during conventional manuscript review less likely. Review articles were disproportionately affected, with a fabrication rate 57% higher than that of other publication types.
The audit also found little evidence of publisher intervention after publication. At the time of analysis, 98.4% of affected papers had received no publisher action, according to the authors.
“When references point to non-existent studies, the evidence they claim to support is fictional,” the authors wrote.
Important Limitations
The findings come with important caveats.
The verification pipeline evaluated only references containing PubMed identifiers. Of 125.6 million structured references identified, approximately 97.1 million (77%) could be verified using the authors’ system; the remaining 23% — including websites, books, and other nonindexed material — were excluded.
The authors also emphasized that the system estimates precision rather than recall, meaning fabricated references that escaped detection were not counted. Some disputed references may also exist in sources outside the four databases searched.
In addition, the early-2026 data reflect only the period from Jan. 1 through Feb. 18, 2026, making those estimates more preliminary than the full-year figures from earlier periods.
Implications for Clinical Literature
For clinicians, the findings raise broader concerns about evidence integrity rather than immediate bedside risk.
Reference inaccuracies in medical publishing are not new; the authors cite prior systematic reviews estimating that roughly one in four references in medical journal articles contains some type of error. What distinguishes fabricated references is that the cited study may not exist at all.
The authors argue that automated reference verification tools already exist and could be incorporated into journal submission workflows before peer review. They also recommend that publishers retrospectively screen existing literature and issue corrections or retractions when fabricated references materially affect a paper’s conclusions.
The Correspondence did not evaluate whether fabricated references changed clinical outcomes or altered treatment recommendations directly. However, the authors note that paper mill publications have previously appeared in systematic reviews used to inform clinical guidelines, potentially weakening the evidence chain underlying medical decision-making.
Source: The Lancet