Research Letter
AI in Medicine
Self-Disclosed Use of AI in Research Submissions to BMJ Journals
Isamme AlFayyad, Maurice P. Zeegers, Lex Bouter, et al
JAMA Published Online: January 28, 2026
doi: 10.1001/jama.2025.25688
Artificial intelligence (AI), including generative AI, has rapidly emerged as a powerful tool in scientific research and publication.1 Current estimates of AI use rely on self-reported surveys with limited samples. These surveys have suggested widespread use of AI tools among researchers, with published estimates ranging from 28% to 76%.2,3 In April 2024, BMJ Group implemented mandatory questions on AI use in its manuscript submission systems.4 Authors’ self-disclosed AI use in research manuscripts submitted to 49 BMJ journals was analyzed to assess frequency of use, the types of AI tools used, tasks assisted by AI, and factors associated with disclosure.
Methods
We performed a cross-sectional study of all research manuscripts (including systematic reviews and meta-analyses) submitted between April 8 and November 6, 2024. We reviewed and classified submitting authors’ free-text disclosures of AI tools based on a function-based taxonomy that grouped tools according to their primary purpose within the manuscript. Disclosure prevalence, types of AI used, and tasks assisted by AI are presented as frequencies and proportions. Binominal logistic regression was used to identify which factors are associated with AI use disclosure. Adjusted odds ratios (ORs) were estimated for the 6 variables included in the regression analysis model (journal acceptance rate, Impact Factor, peer review model, journal specialty, region of the submitting author, and number of authors). This study was approved by the research ethics committee of Maastricht University (FHML-REC/2025/014), preregistered in Open Science Framework, and followed the Strengthening the Reporting of Observational Studies in Epidemiology (STROBE) reporting guidelines. A 2-sided P value of .05 was considered statistically significant (eMaterials in Supplement 1).
Results
Among 25 114 eligible submissions, 1431 (5.7%) reported AI use. The most frequently disclosed AI tools were chatbots (56.7%)—with ChatGPT dominating as the tool of choice—followed by writing assistants (12.7%), primarily Grammarly. The specific AI tool used was not disclosed for 449 submissions (31.4%) (Table 1).
Table 1. Reported Prevalence of Artificial Intelligence (AI) Tools Used Among Submissions to BMJ Journals
| Type | No. (%)a |
|---|---|
| Chatbots | 812 (56.7) |
| Writing assistant tools | 182 (12.7) |
| Visual/image processing tools | 38 (2.7) |
| Evidence synthesis tools | 28 (2) |
| Predictive analytics models | 21 (1.5) |
| AI-powered translators | 19 (1.3) |
| AI-powered data analysis tools | 9 (0.6) |
| AI-powered markdown editors | 7 (0.5) |
| Automatic speech recognition systems | 4 (0.3) |
| Other | 7 (0.5) |
| Not disclosed | 449 (31.4) |
Most authors (87.2%) reported using AI to improve the quality of writing. The findings show an increase in AI use disclosure from 121 of 2693 (4.5%) in April to 286 of 3894 (7.3%) in October. The difference in AI use disclosure across journal specialties was not statistically significant (P = .06) nor between The BMJ and other BMJ journals (P = .32). Authors from South America (OR, 1.75 [95% CI, 1.22-2.49]) and Europe (OR, 1.28 [95% CI, 1.14-1.45]) were more likely to report AI use than those from Asia (Table 2).
Table 2. Descriptive and Binominal Logistic Regression Analysis of Submissions to BMJ Journals Disclosing Artificial Intelligence (AI) Use vs Not Disclosing AI Usea
| Characteristic | No. (%) | Adjusted OR (95% CI) | ||
|---|---|---|---|---|
| Total submissions (N = 25 114) | AI use not disclosed (n = 23 683) | AI use disclosed (n = 1431) | ||
| Journal acceptance | 0.99 (0.99-1.01) | |||
| Mean (SD) | 25.6 (11.2) | 25.6 (11.2) | 25.6 (11.2) | |
| Median (IQR) | 27 (14-36) | 27 (14-36) | 27 (14-36) | |
| No. of authors | 0.99 (0.98-0.99)b | |||
| Mean (SD) | 8.9 (6.5) | 8.9 (6.4) | 8.4 (6.5) | |
| Median (IQR) | 7 (5-11) | 7 (5-11) | 7 (5-10) | |
| Journal Impact Factor | ||||
| None (n = 4) | 51 (2.0) | 47 (0.2) | 4 (0.3) | 1.62 (0.52-5.01) |
| <5 (n = 28) | 14 856 (59.2) | 14 018 (59.2) | 838 (58.6) | 0.88 (0.70-1.11) |
| 5-10 (n = 9) | 4535 (18.1) | 4294 (18.1) | 241 (16.8) | 0.84 (0.68-1.04) |
| 11-20 (n = 1) | 2039 (8.1) | 1916 (8.1) | 123 (7.6) | 1.03 (0.80-1.33) |
| >20 (n = 3) | 3633 (14.5) | 3408 (14.4) | 225 (15.7) | 1 [Reference] |
| Region of submitting author | ||||
| Europe | 6523 (26.0) | 6080 (25.7) | 443 (31.0) | 1.28 (1.14-1.45)b |
| North America | 2795 (11.1) | 2671 (11.3) | 124 (8.7) | 0.81 (0.67-0.97)b |
| Africa | 1196 (4.8) | 1119 (4.7) | 77 (5.4) | 1.17 (0.91-1.50) |
| Oceania | 708 (2.8) | 694 (2.9) | 14 (1.0) | 0.35 (0.21-0.60)b |
| South America | 387 (1.5) | 352 (1.5) | 35 (2.4) | 1.75 (1.22-2.49)b |
| Asia | 13 505 (53.8) | 12 767 (53.9) | 738 (51.6) | 1 [Reference] |
| Peer review model | ||||
| Open peer review (n = 5) | 8600 (34.2) | 8103 (34.2) | 497 (34.7) | 1.02 (0.72-1.46) |
| Single anonymized (n = 36) | 15 289 (60.9) | 14 415 (60.9) | 874 (61.1) | 1.06 (0.80-1.40) |
| Double anonymized (n = 4) | 1225 (4.9) | 1165 (4.9) | 60 (4.2) | 1 [Reference] |
| Journal specialty | ||||
| General medical journal (n = 5) | 10 256 (40.8) | 9654 (40.8) | 602 (42.1) | 1.07 (0.87-1.32) |
| Specialized medical journal (n = 40) | 14 858 (59.2) | 14 029 (59.2) | 829 (57.9) | 1 [Reference] |
Discussion
Overall, 5.7% of authors submitting research manuscripts to BMJ journals disclosed AI use, which is substantially lower than estimates from recent surveys of researchers’ general use of AI.2,3 However, this finding aligns with a similar study of more than 100 000 manuscripts submitted to JAMA and JAMA Network journals that reported that 3.3% of manuscripts contained declarations of AI use.5 Taken together, these results suggest underreporting, perhaps reflecting authors’ deliberate omission or uncertainty over what AI use needs to be disclosed.6
Authors from South America and Europe were significantly more likely to report AI use than those from Asia, which may reflect differences in reporting culture, AI access and training, or local institutional policies.
This study provided real-world data from 49 BMJ journals representing diverse global regions. However, the findings may not generalize to other publishers with different editorial policies or journal scopes. This study is based on data from 2024, which may not accurately reflect the rapidly evolving landscape of AI use. Additionally, submitting authors may not have been fully aware of potential AI use by their coauthors, introducing the possibility of underreporting.
Current mandatory questions the BMJ Group poses on AI use may have only limited value and other measures are needed to improve declaration and support detection of undisclosed use.
