Research Letter
AI in Medicine
Author Disclosure of Use of AI in Submissions to 13 JAMA Network Journals
Roy H. Perlis, Annette Flanagin, Jacob Kendall-Taylor, et al
JAMA Published Online: January 28, 2026
doi: 10.1001/jama.2025.25300
The scientific community has not reached consensus about the appropriate application of artificial intelligence (AI) in manuscript preparation and review.1 In August 2023, JAMA Network journals began requiring authors to disclose use of AI in the preparation of submitted manuscripts.2 We examined these self-reported data to inform policy discussions regarding when and how such AI use should be permitted, guided, and reported.
Methods
This was a cross-sectional study of all manuscripts submitted to 13 JAMA Network journals between August 29, 2023 (after AI questions were deployed), and October 31, 2025. Reason for AI use was extracted from author-reported free-text fields in the submission system using a secure large language model (eAppendix in Supplement 1). Counts and proportions of manuscripts with author-declared use of AI, reasons for AI use, and factors associated with declared AI use were examined. Linear regression was used to examine change in proportion of articles with author-reported use of AI by month. Multivariable logistic regression was used to examine associations between AI use and manuscript characteristics (journal, manuscript type, manuscript disposition, and whether or not the corresponding author’s country of origin considers English as an official or de facto official language, as defined by the World Factbook,3 with an additional term for days since August 29, 2023, to account for secular trend). An uncorrected 2-sided P ≤ .05 was considered the threshold for statistical significance. This study was deemed exempt from approval by the Mass General Brigham Institutional Review Board and used the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guideline.
Results
In total, 105 538 manuscripts were submitted during the 27-month study period, and 3459 (3.3%) declared use of AI; use increased significantly during the study period from 1.71% (95% CI, 1.27%-2.15%) to 5.97% (95% CI, 5.28%-6.66%) (slope, +0.20% [95% CI, 0.16%-0.24%] per month) (Figure 1). AI uses included correction or refinement of language (67.7%), statistical model development (7.3%), other data analysis (6.3%), manuscript drafting (5.5%), and for search and evaluation of literature (4.3%), while 8.8% were not classified in any of these categories. Manuscript characteristics and associations with declared use of AI are reported in Figure 2. Characteristics associated with greater likelihood of author-reported AI use included being Viewpoints (odds ratio [OR], 1.78; 95% CI, 1.56-2.03) or Letters to the Editor (OR, 1.72; 95% CI, 1.50-1.97) vs Original Investigations; corresponding authorship from a country without English as an official language (OR, 1.30; 95% CI, 1.20-1.40); and manuscripts that were rejected (OR, 1.29; 95% CI, 1.05-1.62) or withdrawn prior to review (OR, 8.48; 95% CI, 5.30-13.22) vs accepted or asked to revise. All specialty journal submissions were less likely to be associated with AI use than JAMA submissions except JAMA Oncology and JAMA Psychiatry. The former was not significantly different from JAMA, while the latter was significantly more likely to be associated with AI use (OR, 1.18; 95% CI, 1.00-1.38).
Figure 1. Timeline of Rates of Author Disclosure of Artificial Intelligence (AI) Use in Manuscripts Submitted to JAMA Network Journals, September 2023 to October 2025
Shaded areas indicate 95% CIs.
Figure 2. Multivariable Logistic Regression of Associations Between Author Disclosure of Use of Artificial Intelligence (AI) and Manuscript Characteristics
Regression model includes all displayed variables as well as days since August 29, 2023.
Discussion
Among manuscripts submitted to 13 JAMA Network journals, author declaration of AI use was relatively low but increased over 27 months to about 6%, most commonly for refinement of language. Manuscripts subsequently withdrawn or rejected were more likely to disclose AI than those ultimately accepted or asked to revise. Absent a reference standard for confirming AI use, the extent to which corresponding authors may underreport use cannot be determined, including circumstances in which other authors use AI without corresponding author awareness. Furthermore, as all journals in this study belonged to a single publisher and discipline, the extent to which these patterns generalize requires further investigation. These results are generally similar to other reports; for example, another study of 49 BMJ Group journals reported that 5.7% of authors reported use of AI in submitted manuscripts in 2024,4 and a study of abstracts and methods sections in manuscripts submitted to 10 cancer journals found AI use at even higher rates in 2024.5 These results suggest that AI use in medical publications is becoming more common, which may necessitate greater effort by journals to confirm appropriate use, accountability, and accuracy.
