Fabricated citations: an audit across 2·5 million biomedical papers
Open accessAbstract
Scientific literature depends on the integrity of its references. Each reference implicitly asserts that a verifiable source exists and supports the claims being made. When references point to non-existent studies, readers, reviewers, and policy makers are unable to evaluate the evidence.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
8 of 9 cited sources checked · 2 supported · 1 flagged · accuracy not yet measured
- Premisecitation§ Fabricated citations: an audit across 2•5 million biomedical papers
Systematic reviews that inform clinical guidelines have included paper mill articles.
“Paper mill articles have been included in systematic reviews informing clinical guidelines; 10”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [10] Tang, 2025, Citation contamination by paper mill articles in systematic reviews of the life sciences, JAMA Netw Open source
The Results (Contamination Prevalence; page unavailable) identify 299 life-sciences systematic reviews that incorporated paper mill articles. The Introduction says systematic reviews inform clinical decision-making, but the paper does not establish that the contaminated reviews informed clinical guidelines specifically. (LLM, conf=0.85)
Experimental — accuracy not yet measured
- Premisecitation§ Fabricated citations: an audit across 2•5 million biomedical papers
LLMs are known to create convincing but non-existent references, a documented failure mode.
“Large language models (LLMs) generate plausible sounding but fictitious references, a well documented failure mode”
Quote in the paperSubstantial claim2 of 2 sources checked- cites [2] Athaluri, 2023, Exploring the boundaries of reality: investigating the phenomenon of artificial intelligence hallucination in scientific writing through ChatGPT references, Cureus source
The Results and Discussion report that ChatGPT generated 28 references that could not be found in an extensive search and had no functional DOI, supporting the fictitious-reference failure mode. The study examines ChatGPT, however, and does not establish that LLMs generally produce such references or assess how plausible they sound. (LLM, conf=0.82)
Experimental — accuracy not yet measured - cites [3] Walters, 2023, Fabrication and errors in the bibliographic citations generated by ChatGPT, Sci Rep source
- Premisecitation§ Fabricated citations: an audit across 2•5 million biomedical papers
Prior systematic reviews report that about 25% of references in medical journals contain errors.
“Systematic reviews have found that approximately one in four references in medical journal articles contains errors, 4”
Quote in the paperSubstantial claim1 of 1 sources checked- cites [4] Jergas, 2015, Quotation accuracy in medical journal articles-a systematic review and meta-analysis, PeerJ source
- Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papers
The rate of fabricated references per 10 000 papers grew more than twelvefold, from about 4 in 2023 to 51.3 in Q4 2025 and 56.9 in early 2026.
“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”
Quote in the paperSubstantial claim - Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papers
Review articles showed a 57% higher fabrication rate than other article types (16.7 vs 10.6 per 10 000; p<0.0001).
“Review articles had a fabrication rate that was 57% higher than other paper types (16•7 per 10 000 vs 10•6 per 10 000; p<0•0001; appendix p 7-8).”
Quote in the paperSubstantial claim
- Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papers
The audit found 4046 fabricated references in 2810 papers among the 97.1 million verified references.
“Among 97•1 million verified r e f e r e n c e s , w e i d e n t i f i e d 4046 fabricated references across 2810 papers”
Quote in the paperSubstantial claim - Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papers
At the time of the audit, 98.4% of the 2810 affected papers had not been acted on by their publisher.
“Of the 2810 affected papers, 98•4% had received no publisher action at the time of our audit (appendix p 7).”
Quote in the paperSubstantial claim - Interpretationcitation§ Fabricated citations: an audit across 2•5 million biomedical papers
Checking references is not a routine part of peer review.
“confirming that reference verification is not standard in peer review.”
Quote in the paperSubstantial claim - Interpretationreasoning§ Fabricated citations: an audit across 2•5 million biomedical papers
If a guideline relies on a paper whose reference list is partly fabricated, the evidence base for treatment decisions is weakened.
“when a guideline cites a paper with a partly fictional reference list, the evidence chain for treatment decisions is compromised.”
Quote in the paperSubstantial claim
- Classified as probably not a substantial claim. 5
- May combine two claims. 12
Probably not a substantial claim: may not pass S3 (load-bearing).
Premisecitation§ Fabricated citations: an audit across 2•5 million biomedical papersPrior work puts the share of fabricated references among LLM-generated biomedical citations at 30-69%.
“previous studies estimate that 30-69% of LLM-generated references in biomedical contexts are fabricated.2,3”
Quote in the paperNot a substantial claim2 of 2 sources checked- cites [2] Athaluri, 2023, Exploring the boundaries of reality: investigating the phenomenon of artificial intelligence hallucination in scientific writing through ChatGPT references, Cureus source
The Results report 28 references that could not be found out of 178 generated, about 16%, not 30–69%. The Discussion distinguishes those fabricated references from 41 real articles whose DOIs were erroneous; the figure 69 is a count of references with DOI problems, not a percentage of fabricated references. (LLM, conf=0.88)
Checked by a reasoning model — may be wrong.
Experimental — accuracy not yet measured - cites [3] Walters, 2023, Fabrication and errors in the bibliographic citations generated by ChatGPT, Sci Rep source
The Introduction’s “Previous research” section reports that 64% of citations in a radiology study were fabricated, supporting the biomedical context. But its stated typical range is 47–69% across studies, not 30–69% for biomedical references specifically. (LLM, conf=0.87)
Experimental — accuracy not yet measured
May combine two claims.
Premisecitation§ Fabricated citations: an audit across 2•5 million biomedical papersSuch references tend to be well formatted, credited to actual scientists, and dated plausibly, so standard peer review often misses them.
“These references are often correctly formatted, attributed to real researchers, and bear plausible publication dates, making them difficult to detect by conventional peer review.4”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [4] Jergas, 2015, Quotation accuracy in medical journal articles-a systematic review and meta-analysis, PeerJ source
The Discussion (‘What may be the reasons for quotation errors’ and ‘What can be done’; page unavailable) addresses misquotation, copied citations, and checks by reviewers. It does not say that suspect references are correctly formatted, name real researchers, bear plausible dates, or evade peer review for those reasons. Searches for fabricated references and those specific features found no supporting passage. (LLM, conf=0.75)
Experimental — accuracy not yet measured
May combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersOne 2025 oncology paper on ureteroileal anastomotic techniques had 18 of 30 verified references (60%) fabricated; these were topic-specific, credited to real urologists, and dated 2023 or 2024.
“A 2025 paper on ureteroileal anastomotic techniques in an open access oncology journal contained 18 (60%) fabricated references of 30 verified; each fabricated reference was tailored to the paper's narrow surgical topic, attributed to real urologists, and bore claimed publication years of 2023 or 2024.8”
Quote in the paperNot a substantial claim1 of 1 sources checked- cites [8] Ren, 2025, Comparative analysis of ureteroileal anastomotic stricture rates: Bricker versus Wallace techniques in ileal conduit urinary diversion—a single-surgeon study with BMI-matched design and long-term follow-up excluding cancer recurrence bias, Front Oncol source
Split into 2 atoms — 0 confirmed, 1 partially supported, 0 disputed, 0 misattributed, 1 unverifiable. Expand the claim to see per-atom verdicts.
Experimental — accuracy not yet measured
May combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersThe audit verified 97.1 million of 125.6 million references (77%) that had a PMID; the other 23%, mostly non-indexed sources such as websites, books, and grey literature, were left out.
“Of 125•6 million references, 97•1 million (77%) carried a PMID and were verified; the remaining 23%, predominantly non-indexed references to websites, books, and grey literature, were excluded.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersThe proportion of papers with at least one fabricated reference rose from about 1 in 2828 in 2023 to 1 in 458 in 2025 and 1 in 277 in the first seven weeks of 2026.
“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.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersAmong affected papers, 91% (n=2564) had one or two fabricated references, while 246 had three or more.
“Most affected papers (91%, n=2564) contained one or two fabricated references; 246 contained three or more.”
Quote in the paperNot a substantial claimMay combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersThe audit found signs of paper mill activity: two authors co-occurred on 11 papers in one surgical journal in 2025, with 15 fabricated references on CRISPR diagnostics, AI-guided nanovaccines, and gut microbiome biomarkers.
“Beyond individual papers, we identified patterns consistent with paper mill activity: the same two authors appeared across 11 papers in a single surgical journal in 2025, with 15 fabricated references covering CRISPR diagnostics, AI-guided nanovaccines, and gut microbiome biomarkers, all sharing a core coauthorship pair.”
Quote in the paperSubstantial claimMay combine two claims.
Findingown data§ Fabricated citations: an audit across 2•5 million biomedical papersThe fabricated references did not look obviously wrong: they were topic-specific, well formatted, credited to real researchers, and had plausible dates.
“The fabricated references we identified were not obviously defective: topically specific, correctly formatted, attributed to real researchers, and beared plausible publication dates.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Interpretationreasoning§ Fabricated citations: an audit across 2•5 million biomedical papersIf a reference points to a study that does not exist, the evidence it supposedly supports is not real.
“When references point to nonexistent studies, the evidence they claim to support is fictional.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim. May combine two claims.
Study design and conductnone given§ Fabricated citations: an audit across 2•5 million biomedical papersThe system detects the problem but does not establish its cause.
“Our system identifies the problem, not its cause.”
Quote in the paperSubstantial claimClassified as probably not a substantial claim.
Interpretationreasoning§ Fabricated citations: an audit across 2•5 million biomedical papersAutomated checking of references can prevent fabricated citations from entering the published literature.
“Routine automated verification can close this gap before fabricated references reach the published record.”
- Also stated in: § Fabricated citations: an audit across 2•5 million biomedical papers · plain
“Automated reference verification can close this gap.”
Quote in the paperSubstantial claim- Also stated in: § Fabricated citations: an audit across 2•5 million biomedical papers · plain
May combine two claims.
Study design and conductown data§ Fabricated citations: an audit across 2•5 million biomedical papersThe system's precision was 91% (Fleiss' κ=0.71) in a masked 500-entry validation by three independent reviewers; the design measures precision, not recall.
“Precision of our automated reference verification system was 91% (Fleiss' κ=0•71, indicating moderate agreement in about seven of every ten cases), measured in a 500-entry masked validation with three independent reviewers; this design estimates precision but not recall.”
Quote in the paperSubstantial claimMay combine two claims.
Interpretationreasoning§ Fabricated citations: an audit across 2•5 million biomedical papersThe steep rise beginning in mid-2024 matches the expected delay after LLMs became widely used, though paper mill activity and indexing changes may also have played a role.
“The sharp inflection in mid-2024 coincides with the expected publication lag following widespread LLM adoption, although increased paper mill activity and changes in journal indexing practices might also have contributed.”
Quote in the paperSubstantial claimMay combine two claims.
Premisereasoning§ Fabricated citations: an audit across 2•5 million biomedical papersBecause LLMs became widely available in late 2022 and 2023, and journals take 100-200 days from submission to publication, LLM-assisted papers would start appearing in PubMed Central around mid-2024.
“LLMs became broadly available in late 2022 and 2023; with submission-to-publication times of 100-200 days, 9 LLM-assisted papers would appear in PubMed Central from mid-2024 onward.”
Quote in the paperSubstantial claim0 of 1 sources checked- cites [9] Andersen, 2021, Time from submission to publication varied widely for biomedical journals: a systematic review, Curr Med Res Opin source
Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.
Study design and conductreasoning§ Fabricated citations: an audit across 2•5 million biomedical papersDropping the 23% of references without PMIDs could bias the estimates either way: fabricated references may be more frequent in non-indexed sources, leading to undercounting, or they may be more common among references with PMIDs.
“The exclusion of 23% of references with no PMIDs could bias estimates in either direction: fabricated references might be more common among non-indexed sources, including grey literature, websites, and books (undercounting), or they can be preferentially associated with PMIDs”
Quote in the paperNot a substantial claim
The paper relies on this without support.
Before this study, no systematic audit of reference integrity across the biomedical literature had been done.
“To our knowledge, no systematic audit of reference integrity across the biomedical literature has been conducted until now.”
The paper relies on this without support.
Major research integrity databases currently lack a separate category for fabricated references.
“fabricated references do not currently exist as a discrete category in major research integrity databases”
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Recent research suggests that the problem of hallucinated and unreliable citations is already widespread and is becoming increasingly prevalent in the published literature and at scientific conferences.
“Recent research suggests that this problem is already widespread and is becoming increasingly prevalent in the published literature and at scientific conferences.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
In “Fabricated citations: an audit across 2·5 million biomedical papers” (page not provided), the authors report fabricated references across 2810 published papers and a rising rate from 2023 to early 2026. The audit concerns biomedical papers, not scientific conferences; its mention of conference proceedings describes a search database, not a finding about citations at conferences. (LLM, conf=0.88)
Experimental — accuracy not yet measuredRecent studies have shown that the problem of hallucinated references occurs both across large-scale publication corpora and within conference proceedings.
“Recent studies have shown that this problem occurs both across large-scale publication corpora and within conference proceedings.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
The paper’s main text (page not supplied) reports finding 4046 fabricated references across 2810 papers in a large-scale biomedical literature audit. It mentions conference proceedings only as material indexed by Google Scholar; it does not report finding fabricated references within conference proceedings. (LLM, conf=0.88)
Experimental — accuracy not yet measuredA large-scale audit published in The Lancet reported a substantial increase in fabricated citations in the biomedical literature.
“A large-scale audit published in The Lancet reported a substantial increase in fabricated citations in the biomedical literature (Topaz et al., 2026)”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
The Lancet study does not prove that every fabricated citation was directly created by an LLM, and it suggests that possible causes include LLM hallucinations, paper mill activity, or intentional manipulation.
“The Lancet study does not prove that every fabricated citation was directly created by an LLM, and it suggests that possible causes include LLM hallucinations, paper mill activity, or intentional manipulation.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
The Lancet study found 4,046 fabricated citations in 2,810 biomedical papers.
“The Lancet study (Topaz et al., 2026) found 4,046 fabricated citations in 2,810 biomedical papers and showed that their prevalence increased from 2023 to 2025/2026.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
Problematic citations can pass editorial or moderation processes.
“At the same time, both studies show that problematic citations can pass editorial or moderation processes.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
The more serious concern is that hallucinated references may appear credible at first glance, escape detection by reviewers and editors, and ultimately undermine the reliability of the scientific literature.
“From this perspective, the issue is not merely that AI tools sometimes generate incorrect references; the more serious concern is that such references may appear credible at first glance, escape detection by reviewers and editors, and ultimately undermine the reliability of the scientific literature (Topaz et al., 2026; Zhao et al., 2026).”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
The prevalence of fabricated references in biomedical papers increased from approximately one in 2,828 papers in 2023 to one in 458 in 2025 and one in 277 in early 2026.
“In 2023, a fabricated reference was identified in approximately one in 2,828 papers; in 2025, this increased to one in 458 papers; and in early 2026, to one in 277 papers.”
From Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do (2026) · cites this paper as Topaz et al., 2026
Cites
Gusenbauer, M., & Haddaway, N. R. (2020) · Research Synthesis Methods · doi:10.1002/jrsm.1378 · Cited by 2,379 · see citing works
OpenAlex: A fully-open index of scholarly works, authors, venues, institutions, and conceptsPreprint
Priem, J., Piwowar, H. A., & Orr, R. (2022) · arXiv.org · doi:10.48550/arxiv.2205.01833 · Cited by 743 · see citing works
Fabrication and errors in the bibliographic citations generated by ChatGPT
Walters, W. H., & Wilder, E. I. (2023) · Scientific Reports · doi:10.1038/s41598-023-41032-5 · Cited by 499 · see citing works
Athaluri, S. A., Manthena, S. V., Kesapragada, V. S. R. K. M., Yarlagadda, V., Dave, T., & Duddumpudi, R. T. S. (2023) · Cureus · doi:10.7759/cureus.37432 · Cited by 403 · see citing works
The fight against fake-paper factories that churn out sham science
Else, H., & Van Noorden, R. (2021) · Nature · doi:10.1038/d41586-021-00733-5 · Cited by 251 · see citing works
Reference coverage analysis of OpenAlex compared to Web of Science and Scopus
Culbert, J. H., Hobert, A., Jahn, N., Haupka, N., Schmidt, M. A., Donner, P., & Mayr, P. (2025) · Scientometrics · doi:10.1007/s11192-025-05293-3 · Cited by 179 · see citing works
Quotation accuracy in medical journal articles—a systematic review and meta-analysis
Jergas, H., & Baethge, C. (2015) · PeerJ · doi:10.7717/peerj.1364 · Cited by 88 · see citing works
Time from submission to publication varied widely for biomedical journals: a systematic review
Andersen, M. Z., Fonnes, S., & Rosenberg, J. (2021) · Current Medical Research and Opinion · doi:10.1080/03007995.2021.1905622 · Cited by 59 · see citing works
Citation Contamination by Paper Mill Articles in Systematic Reviews of the Life Sciences
Tang, G., & Cai, H. (2025) · JAMA Network Open · doi:10.1001/jamanetworkopen.2025.15160 · Cited by 25 · see citing works
Ren, C., Xiao, M., Zhu, J., Tong, W., & Yi, F. (2025) · Frontiers in Oncology · doi:10.3389/fonc.2025.1613772 · Cited by 1 · see citing works
Flagged first, then by citations
Cited by
Sorted by citations
Updated Guidance for Author Use of AI in Medical Publication
Flanagin, A., Perlis, R. H., & Bibbins-Domingo, K. B. (2026) · JAMA · doi:10.1001/jama.2026.16613 · Cited by 13 · see citing works
False references, real consequences: safeguarding spine evidence
Samartzis, D., Berjano, P. L., Edwards, K., et al. (2026) · European Spine Journal · doi:10.1007/s00586-026-10052-w · Cited by 2 · see citing works
Moons, P. (2026) · European Journal of Cardiovascular Nursing · doi:10.1093/eurjcn/zvag148 · Cited by 2 · see citing works
Detecting Hallucinated and Suspicious Citations: What Current Tools Can and Cannot Do
Badalova, F., & Mayr, P. (2026) · arXiv · doi:10.48550/arxiv.2607.22693 · Cited by 1 · see citing works
“A large-scale audit published in Lancet reported a substantial increase in fabricated citations in the biomedical literature [22], while the large-scale study by Zhao and colleagues demonstrated that non-existent references occur systematically across different publication corpora [26].”
InfluentialAI hallucinations in academic writing: implications for research integrity
Giray, L., Islam, M. E., & Glo, J. A. (2026) · Naunyn-Schmiedeberg s Archives of Pharmacology · doi:10.1007/s00210-026-05494-4 · Cited by 6 · see citing works
Grunebaum, A. N., Dudenhausen, J., & Chervenak, F. (2026) · American Journal of Obstetrics and Gynecology · doi:10.1016/j.ajog.2026.06.011 · Cited by 3 · see citing works
Han, C., Manoharan, S., Ye, X., & Speidel, U. (2026) · Journal of the Association for Information Science and Technology · doi:10.1002/asi.70104 · Cited by 1 · see citing works
“Under this scenario, even if the hallucination rate per generated citation decreases, a higher volume of AI-assisted citation generation could still lead to an increase in the proportion of papers containing unverifiable references (Topaz et al., 2026).”
InfluentialA key science publishing platform is cracking down on AI slop
Kovanović, V. (2026) · doi:10.64628/aa.gpmnkgxkm · Cited by 1 · see citing works
New ‘AI scientists’ are improving – but reveal their fundamental limits
Verspoor, K. M. (2026) · doi:10.64628/aa.994m56a9e · Cited by 1 · see citing works
Gaillard, F., & Jones, J. C. (2012) · Radiopaedia.org · doi:10.53347/rid-19444
Surge in fake citations uncovered by audit of 2.5 million biomedical-science papers
Naddaf, M. (2026) · Nature · doi:10.1038/d41586-026-00748-w
Fagerberg, P., Sallander, O., Patil, K. V., Thunborg, C., Lundstrom, L., Berg, A., Nyman, A., Borg, N., & Linden, T. K. A. (2026) · medRxiv · doi:10.64898/2026.06.02.26354724
An AI-Powered Trisomy 21 Research AssistantPreprint
Nandi, S., Sundararajan, Z., Subirana-Granés, M., Espinosa, J. M., Pividori, M., Sullivan, K. D., Galbraith, M. D., & Costello, J. C. (2026) · bioRxiv (Cold Spring Harbor Laboratory) · doi:10.64898/2026.06.08.730893
From AI Agents to AI Scientists in Health: Emerging Landscape, Challenges, and the Path ForwardPreprint
Chen, Q., Wang, X., Zhou, R., Kim, H., Wang, S., Zhao, W., Liu, T., Riaz, I., He, L., Zhao, Y., et al. (2026) · SSRN Electronic Journal · doi:10.2139/ssrn.6889798 · Cited by 1 · see citing works
Survey evidence demands professional guidance on large language model use in medical physics
Norris, S. A., Kron, T., Douglass, M. J. J., & Badawy, M. K. (2026) · Physica Medica · doi:10.1016/j.ejmp.2026.105835
Grumezescu, A. M. (2026) · Biointerface Research in Applied Chemistry · doi:10.33263/briac164.101
Kharchenko, V., Korneyev, V. A., & Filimonova, N. S. (2026) · Zoodiversity · doi:10.15407/zoo2026.03.301
Mohakud, N. K., & Acharya, G. C. (2026) · Journal of Integrative Medicine and Research · doi:10.4103/jimr.jimr_83_26
Фантомні посилання та криза рецензування: як штучний інтелект перевіряє стійкість наукових видань
Kharchenko, V., Korneyev, V. A., & Філімонова, Н. С. (2026) · Visnik Nacional noi academii nauk Ukrai ni · doi:10.15407/visn2026.06.078
Mehta, N., Benjamin, J. C., Mehta, S., Valanci, S., MacNeill, H., & Masters, K. (2026) · Medical Teacher · doi:10.1080/0142159x.2026.2681971
Bolo-Varela, O., & Pérez, R. C. (2026) · Desde el Sur Revista de Ciencias Humanas y Sociales de la Universidad Científica del Sur · doi:10.21142/des-1802-2026-e0028
Surfacing the Invisible: AI and the IS Discipline
Agarwal, R., Padmanabhan, B., & Liu, Y. (2026) · ACM Transactions on Management Information Systems · doi:10.1145/3828556
“…scientific field and discipline computer science, medicine, astrophysics, organization science [Gartenberg et al. 2026; Si et al. 2025; Sokol 2026; Topaz et al. 2026] finds itself at an inflection point, acknowledging a pressing need for community debate, introspection, and a shared, collective…”
Weiniger, C. F., & Landau, R. (2026) · International Journal of Obstetric Anesthesia · doi:10.1016/j.ijoa.2026.105249
Fabricated references are twice as common in medRxiv preprints as in peer‐reviewed articles
Topaz, M., Zhang, Z., Gupta, P., Roguin, N., & Peltonen, L. (2026) · Journal of Internal Medicine · doi:10.1111/joim.70139
“Our recent audit of 2.5 million PubMed Central (PMC) articles iden-tified 4046 fabricated references, with the rate growing more than 12-fold from 2023 to early 2026 [2].”
InfluentialSelf-citationBryant, A. (2026) · Informatics · doi:10.3390/informatics13070113
AI Use in Orthopedic Publishing: A Shared Accountability
Jacobson, J., & Cornell, C. N. (2026) · HSS Journal® The Musculoskeletal Journal of Hospital for Special Surgery · doi:10.1177/15563316261469568
The rise of AI slop in transfusion medicine: Urgent need for oversight
Raza, S., Booth, G. S., & Jacobs, J. W. (2026) · Transfusion · doi:10.1111/trf.70347
Artificial Intelligence and Accountability in Scientific Writing: The Author’s Responsibility
Akgün, M. (2026) · Thoracic research and practice · doi:10.4274/thoracrespract.2026.2026.100726
Silva, A., & Silva, V. S. E. (2026) · European Journal of Cancer · doi:10.1016/j.ejca.2026.116963
Notley, C. J. (2026) · Nicotine & Tobacco Research · doi:10.1093/ntr/ntag160
Let's use ChatGPT for research: fictitious citations, who's responsible?
Abdullah, S., Anwar, F., Khalid, Z., & Siddique, A. (2026) · Consortium Psychiatricum · doi:10.17816/cp15879
Record, S. M., Giannas, E., Kokosis, G., et al. (2026) · Annals of Plastic Surgery · doi:10.1097/sap.0000000000004851
El proceso editorial y la inteligencia artificial (segunda parte)
Acosta, L. E. R. (2026) · Revista Facultad de Ciencias Económicas · doi:10.18359/rfce.8731
Citation hallucination, model collapse, and framework of clinical decision making
Allareddy, V., Allareddy, V., Rampa, S., & Romesh, N. (2026) · BDJ · doi:10.1038/s41415-026-0094-9
When AI agrees with us: large language models as amplifiers of scientific bias
Barber, L. A., Chen, V., Kang, E. C., Yuan, V., Mell, S. P., Cecchinato, R., Samartzis, D., & Sima, S. (2026) · European Spine Journal · doi:10.1007/s00586-026-10289-5
Moons, P. (2026) · European Journal of Cardiovascular Nursing · doi:10.1093/eurjcn/zvag178
Saravanan, S., Chan, S. C. C., & Rashid, A. (2026) · Medical Teacher · doi:10.1080/0142159x.2026.2712427
Shrime, M. G. (2026) · BMJ Global Health · doi:10.1136/bmjgh-2026-025873
The Ethics of Prompting Large Language Models in Academic Publishing
Daly, T. P., & da Silva, J. A. T. (2026) · Annals of Biomedical Engineering · doi:10.1007/s10439-026-04356-7 · Cited by 1 · see citing works
A Conversation With … Kent R. Anderson MBA, Veteran of Medical Publishing, on What Readers Can Trust
Leopold, S. S. (2026) · Clinical Orthopaedics and Related Research · doi:10.1097/corr.0000000000004113
The golden thread of AI subalternity: epistemic activism
Garcia, I. K. (2026) · AI and Ethics · doi:10.1007/s43681-026-01318-6 · Cited by 1 · see citing works
“…of the material issues that AI subalterns are combatting, including the aggravation of climate change through AI-dedicated data centres, the theft of intellectual property [16 ], the corrosion of scientific publications [ 95], the rise of “AI psychosis” [100], and the deaths of many people [5].”
Reference quality of OpenEvidence across five medical specialties
Lotan, T. L., Omar, M., Barash, Y., Kruskal, J. B., Ahmed, M., Brook, O. R., Klang, E., & Gorenshtein, A. (2026) · npj Health Systems · doi:10.1038/s44401-026-00142-8 · Cited by 1 · see citing works
“A separate audit of 2.5 million published papers found fabrication rates rising 12-fold between 2023 and early 2026 5 .”
InfluentialAI Slop in Radiology: Cutting Through the Clutter
Bera, K., Ramaiya, N. H., & Tirumani, S. H. (2026) · American Journal of Roentgenology · doi:10.2214/ajr.26.35537
Nations expanding their research programmes impose sweeping penalties for malpractice
Plackett, B. (2026) · Nature · doi:10.1038/d41586-026-02517-1
Shin, R., Lee, J.-M., Park, J., Kwun, J.-S., Cho, H.-W., Kang, S., & Jeon, K. (2026) · medRxiv · doi:10.64898/2026.08.31.26361806
Brenner, A., & Müller, A. (2026) · Nurse Education in Practice · doi:10.1016/j.nepr.2026.104969
Jung, E., Yoo, J., Kim, S. G., & Kim, Y. S. (2026) · Gastroenterology report · doi:10.1093/gastro/goag098
Authors Must Have Intellectual Accountability and Thus Cannot Be ‘Brain-Dead’
Tang, B. L. (2026) · Medical Ethics and Bioethics · doi:10.3390/meb1010002
“…and responsible AI use would be beneficial, one also needs to be wary of potential cognitive deskilling and AI-dependency [24], as well as the lure of cutting corners with AI, which authors of the thousands of biomedical papers with AI-fabricated references [25] might now live to regret.”
Ponce, C. J. (2026) · Observatorio Económico · doi:10.11565/oe.v1i214.629
Bartz, A., Glaser, N., Bayer, O., Hallsson, L. R., Jaeger, V. K., Krumkamp, R., Minnerup, H., Pickford, R., Olsen, K. S., Wülk, N., et al. (2026) · Research Square · doi:10.21203/rs.3.rs-10974790/v1
Toward autonomous science with agentic artificial intelligence
Guo, L. Y., Ting, D. S. J., Su, X., Aliper, A., Zhavoronkov, A. A., & Ting, D. S. W. (2026) · Cell · doi:10.1016/j.cell.2026.08.052 · Cited by 2 · see citing works
Guellec, B. L., Bentegeac, R., Boyer, L., Celi, L. A., Boussat, B., & Hamroun, A. (2026) · Journal of Epidemiology and Population Health · doi:10.1016/j.jeph.2026.203798 · Cited by 1 · see citing works
Basra, R. (2026) · International Journal of Modern Science and Research Technology · doi:10.67317/ijmsrt26sep073
Beyond the Black Box: A Point-Of-Care Framework for LLM Literacy and Critical Appraisal
Stephenson, C. R., Aakre, C. A., Croghan, I. T., Hurt, R. T., & West, C. P. (2026) · Journal of Primary Care & Community Health · doi:10.1177/21501319261491056
Ng, J. Y., Tan, J., Syed, N., Adapa, K., Gupta, P. K., Li, S., Mehta, D. H., Ring, M. R., Shridhar, M., Souza, J. P., et al. (2026) · Journal of Evidence-Based Medicine · doi:10.1111/jebm.70189 · Cited by 1 · see citing works
Self-citationHart, M. L., Walsh, M. C., Raimondo, C., Kato, R., & Rolauffs, B. (2026) · Frontiers in Bioengineering and Biotechnology · doi:10.3389/fbioe.2026.1933082
“…concern is the capacity of AI platforms to inadvertently incorporate fraudulent, “paper-mill” generated, or retracted literature into knowledge synthesis (Abalkina et al., 2025; Else and Van Noorden, 2021; Kendall and Teixeira da Silva, 2024; Kobak et al., 2025; Thelwall, 2025; Topaz et al., 2026).”
Fabricated citations at scale: from detection to prevention
Li, T., Nguyen, D.-D., & Ross, J. S. (2026) · The Lancet · doi:10.1016/s0140-6736(26)01434-0
Fabricated citations at scale: from detection to prevention
Mousa, A. M., & Mahmoud-Katheer, M. (2026) · The Lancet · doi:10.1016/s0140-6736(26)01435-2
Uso Responsable de la Inteligencia Artificial en las Publicaciones Médicas
Herrera, M. A. O. (2026) · Revista Guatemalteca de Urología · doi:10.54212/27068048.v14i1.208
Fabricated citations at scale: from detection to prevention – Authors' reply
Topaz, M., Roguin, N., Gupta, P., et al. (2026) · The Lancet · doi:10.1016/s0140-6736(26)01589-8
Self-citationA pragmatic workflow for integrating large language models into toxicological risk assessment
Bassan, A., Amberg, A., Barreca, M. L., Bradley, P. M., Bringezu, F., Souza, J. D. P., Guicheney, C., Piparo, E. L., Kovarich, S., Massarsky, A., et al. (2026) · Frontiers in Toxicology · doi:10.3389/ftox.2026.1877128
Scientific Writing, Peer Review, and Artificial Intelligence
BaHammam, A. S. (2026) · Sleep Medicine Clinics · doi:10.1016/j.jsmc.2026.08.004
Nurse Educators’ Use of AI Tools in Scholarly Writing
Oermann, M. H. (2026) · Nurse Educator · doi:10.1097/nne.0000000000002340
Medvedev, D. A. (2026) · Science Editor and Publisher · doi:10.24069/sep.2026.11.48
Krumholz, H. M. (2026) · Journal of the American College of Cardiology · doi:10.1016/j.jacc.2026.08.032
Miranović, T., & Mehta, P. (2026) · Oxford University Press eBooks · doi:10.1093/9780198945215.003.0243
Nam, Y., Jeong, J., & Kim, N. (2026) · arXiv · doi:10.48550/arxiv.2606.09500
- Authors
- Topaz, M., Roguin, N., Gupta, P., Zhang, Z., & Peltonen, L.-M.
- Affiliations
- VNS Health (US)
- Ben-Gurion University of the Negev (IL)
- Columbia University (US)
- University of Turku (FI)
- Corresponding author
- Topaz, Maxim
- Venue
- The Lancet
- Publisher
- Elsevier BV
- Published
- 2026-05
- Volume
- 407
- Issue
- 10541
- Pages
- 1779-1781
- ISSN
- 0099-5355 (ISSN-L)
- 0140-6736
- OpenAlex
- W7160490639
- Cited by
- 85
- References
- 8
- Citations by year
- 2026: 84 · 2012: 1
- FWCI
- 324.04
- Type
- journal article
- Language
- en
- Update notices
- erratum (2026-07-18)
- Open access
- hybrid
- doi.org
- Licence
- Locations
- The Lancet (publishedVersion, other-oa)
- PubMed (publishedVersion)
- Topics
- Social Sciences › Safety Research › Academic integrity and plagiarism
- Social Sciences › Decision Sciences › Statistics, Probability and Uncertainty › scientometrics and bibliometrics research
- Social Sciences › Decision Sciences › Information Systems and Management › Academic Publishing and Open Access
- Keywords
- citation integrity, reference verification, biomedical literature, research integrity
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-06.
Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-06.
Cited sources are checked by gpt-6-sol.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
