Our Take
The data is solid but masks a deeper crisis: researchers are using AI as a shortcut to avoid actually reading the papers they cite.
Why it matters
These fabricated citations corrupt systematic reviews and clinical guidelines that rely on cited studies. The trend signals that AI tools are making academic engagement with literature increasingly superficial.
Do this week
Researchers: audit your last three papers for citation accuracy and implement manual verification of all AI-suggested references before submission.
Fabricated citations increased sixfold in two years
Columbia University researchers analyzed over 2 million papers and 97 million citations using AI tools, identifying approximately 4,000 fabricated citations across 2,800 papers (per the Lancet study). The frequency jumped from 1 in 2,828 papers containing fake citations in 2023 to 1 in 458 papers in 2025. During the first seven weeks of 2026, the rate reached 1 in 277 papers.
The study was triggered when lead researcher Maxim Topaz discovered a fabricated citation in his own work after using an AI chatbot for editing assistance. Despite checking citations, an erroneous reference nearly made it to publication. "I was deeply embarrassed: I checked for that, and it still almost happened to me," Topaz said.
The fabricated citations concentrate heavily in two large open-access publishers, which account for more than a third of all fake references (publishers unnamed in the study). These publishers charge authors high fees to publish papers without paywalls.
Citation culture is shifting from engagement to automation
The fabricated citations directly threaten systematic reviews and clinical guidelines that build on cited studies. But the deeper issue is changing research practices. Researchers now prompt AI tools based on hunches rather than reading papers and taking reflective notes, according to Northwestern University's Mohammad Hosseini.
This represents a shift from citations as genuine signals of engagement with literature to box-checking exercises. The presence of hallucinated citations "shows that there's people who don't even want to spend half an hour to check the references of a paper," Hosseini noted, reflecting desperation to publish quickly in a flawed academic evaluation system.
Publishers are implementing automated detection
Major publishers are responding with varying approaches. Science family journals use automated reference checking tools and report no published papers with fabricated citations. The New England Journal of Medicine and JAMA have similar validation tools, with authors contractually responsible for their work.
The Public Library of Science reports "numerous" unverifiable references in submissions and is piloting detection tools. However, they note high false positive rates due to formatting issues, incomplete information, and database limitations that flag legitimate references as fabricated.
For working researchers, the solution remains manual: verify every AI-suggested citation against the actual source before submission. The time saved by AI assistance gets consumed by the verification step that responsible research has always required.