AI evidence extraction

The Presence and Nature of AI-Use Disclosure Statements in Medical Education Journals: A Bibliometric Study

Lauren A. Maggio, Hamzah S. Algodi, Joe A. Costello, Erik W. Driessen, Kevin Oswald, Lorelei Lingard · Perspectives on Medical Education · 2026

AI-generated evidence extraction, verified across multiple analytical personas. Not a substitute for the peer-reviewed original.

9/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

14.22

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.5334/pme.2431

Methodology & findings

Study design

Bibliometric study using structured screening and data extraction.

Sample

N = 2046, 7 groups

Primary method

Descriptive statistics to characterize bibliographic and author-level variables. Descriptive content analysis for AI tool used, nature of usage, and responsibility attestation. Two researchers independently analyzed subsets of disclosures with regular meetings to discuss and resolve discrepancies.

Main result

The study found that "only 2.5% (n = 51) had an AI disclosure statement" among 2,046 empirical articles published in 24 leading medical education journals during the first half of 2025. Additionally, "Use of AI for editing purposes was the most reported (n = 28, 54.9%), followed by transcription (n = 12, 23.5%), thematic/data analysis (n = 7, 13.7%), drafting (n = 3, 5.9%), and article screening in knowledge syntheses (n = 2, 3.9%)." The research also found that "Only 26% (n = 13) of disclosures included an attestation of responsibility."

Reports effect sizes.

Research paradigm

Empirical positivism with descriptive quantitative analysis

Author conclusions

The authors conclude: "AI-use disclosures in medical education journals are rare and appear mostly in work from non-native English-speaking regions of the world. A shared disclosure practice is evident: name the tool and affirm author responsibility, but describe use superficially. This suggests a practice of 'safe' disclosure that may fail to satisfy the goal of ensuring transparent and ethical AI use in research." They also note that "the combination of low disclosure rates and safe disclosure practice produces a 'transparency paradox' in which mandatory disclosure that does not attend to social complexities leads to both nondisclosed AI use, and disclosure 'theatre' in which published disclosures are more performative than informative."

Risk of bias

Selection bias: Limited to 24 specific medical education journals; non-empirical articles excluded; Temporal bias: Only 6-month sampling period; publication lag of approximately 188 days may mean recent submissions not yet published; Information bias: Instructions reviewed at time of data collection may differ from guidance when authors conducted work; Detection bias: AI use detection limited to explicit disclosures; undisclosed AI use not captured; Selection bias: Analysis limited to 6-month period, which may not be representative of full-year patterns or trends over time; Sampling frame bias: Limited to 24 leading medical education journals; smaller or emerging journals not included; Detection bias: Reliance on explicit disclosure statements may miss undisclosed AI use; study does not validate accuracy of disclosed AI use; Language bias: Focus on English-language medical education journals; Publication lag bias: Average 188-day lag from submission to publication means manuscripts referenced by editors may not yet have been published during study period; Selection bias: Limited to 6-month sampling period, may not capture temporal trends or seasonal variations in publication practices; Measurement bias: Analysis limited to first authors only; corresponding authors and other contributors not assessed; Database bias: Web of Science used as primary source for author metrics; some author publications may not be indexed; Publication lag bias: Average 188-day lag from submission to publication may mean manuscripts referenced by editors were not yet published during study period; Language bias: Study focused on journals in English; non-English journals excluded; Tool detection bias: Study relies on explicit disclosure; undisclosed AI use not captured

Limitations

The authors state: "First, our analysis was limited to the first 6 months of 2025
Because editors have reported a recent increase in AI-related disclosures, it is possible that disclosure rates were higher during this period, but we lack data to examine changes or trends over time." Additionally, "Second, we did not compare articles with AI disclosures to those without, which limits our ability to assess differences in overall transparency practices." Furthermore, "Finally, because many journals impose word limits, authors may have a limited word count to include and/or elaborate on their AI use, potentially constraining the detail and desire to include a disclosure statement."

Open questions raised

Need to explore underpinning influences on medical education field's current rate of AI-use disclosure
Future research should explore how authors are deciding what and how to disclose AI use
Need for investigation of differences in overall transparency practices between articles with and without AI disclosures
Examination of temporal trends in AI-use disclosure over time
Study of equity implications of AI-use disclosure practices and policies
Need for longitudinal analysis to examine temporal trends in AI-use disclosure rates

Extracted from: pdfAgreement 69%

Explore related topics

Responsible AI Use Research Integrity Content Analysis