AI evidence extraction

Emerging conventions in GenAI disclosure: how applied linguistics scholars disclose AI use

Benjamin Luke Moorhouse, Hassan Nejadghanbar, Chenze Wu, Harsharan Kaur, Marie Alina Yeo · Applied Linguistics Review · 2026

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

9/10

Relevance

3/4

Quality (LMQS)

Evidence

Citations

7.16

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.1515/applirev-2025-0291

Methodology & findings

Study design

Systematic content analysis of formal disclosure sections (acknowledgements, declarations, additional information) in published research articles.

Main result

Only 4.2% of the 3,384 articles examined included any disclosure related to GenAI use, with 2.8% declaring use and 1.4% declaring non-use. The study found that "of the 94 articles that explicitly reported GenAI use, the vast majority of statements (f = 86) mention the specific tool used (e.g., ChatGPT, Grammarly), however, fewer authors included the specific version (e.g., GPT.4o) (f = 30)." The most common reported uses of AI are "language editing improvement/refinement (including checking grammar and spelling) (f = 48)" and "enhancing readability/clarity (f = 40)." When disclosures appear, "most identify the tool (most commonly ChatGPT or Grammarly) and its general purpose (language editing, readability, proofreading)."

Research paradigm

Interpretivism/Qualitative empiricism with descriptive quantification

Author conclusions

The authors conclude: "According to our study, the applied linguistics publishing field appears to be transitioning from ad hoc acknowledgements to more structured, policy-led disclosures of GenAI use in research and writing for publication. At present, statements often seem to satisfy transparency requirements without fully enabling evaluative insight. A shift to concise, structured, and functionally placed disclosures, supported by GenAI-LR development and clear editorial policies, could help the field move beyond performative transparency toward practical accountability."

Risk of bias

Selection bias: Only SSCI-indexed journals included; excludes non-English language journals and non-indexed journals; Measurement bias: Coding captures declared practice rather than actual practice; relies on author self-reporting; Underreporting bias: Authors may underreport or not report GenAI use despite actual use due to stigma; Language bias: Exclusion of non-English journals due to researcher linguistic limitations; Journal prestige bias: Focus on high-impact journals may not reflect broader field practices; Selection bias: Exclusion of non-English journals limits generalizability to multilingual scholarship; Selection bias: Focus only on SSCI-indexed journals excludes lower-tier but potentially influential journals; Detection bias: Coding captures declared practice rather than actual practice; under-reporting of GenAI use likely due to stigma; Methodological bias: Methods sections not systematically analyzed, potentially missing important methodological disclosures; Researcher bias: Five-person team coding with potential subjective interpretation despite standardization procedures; Selection bias: Study limited to SSCI-indexed journals only, excluding other publication venues and languages; Reporting bias: Authors acknowledge that 'under-reporting is plausible' and actual GenAI use likely exceeds declared use; Detection bias: Study captures only declared practices in formal disclosure sections, not actual practices or disclosures in Methods sections; Language bias: Exclusion of non-English journals due to research team linguistic competence; Journal policy bias: Disclosure rates highly sensitive to journal policies, with mandatory policies showing 100% reporting (ReCALL) versus sporadic disclosures in other journals

Open questions raised

Future research should (a) sample Methods sections of articles to capture methodological disclosures, (b) experimentally test how disclosure content and placement affect readers' perceptions of trust, rigor, and credibility, and (c) investigate authors' decision-making regarding whether and how to disclose, including perceived risks and benefits across sub-fields and author demographics.
Future research should: (a) sample Methods sections of articles to capture methodological disclosures, (b) experimentally test how disclosure content and placement affect readers' perceptions of trust, rigor, and credibility, and (c) investigate authors' decision-making regarding whether and how to disclose, including perceived risks and benefits across sub-fields and author demographics.
The authors identify three directions for future research: (a) sample Methods sections of articles to capture methodological disclosures; (b) experimentally test how disclosure content and placement affect readers' perceptions of trust, rigor, and credibility; and (c) investigate authors' decision-making regarding whether and how to disclose, including perceived risks and benefits across sub-fields and author demographics.

Data: No explicitly available datasets are mentioned. The authors note that 'a total of 3,384 articles were extracted' from 84 journals, but no public dataset repository is provided.Code: No code repositories mentioned.Extracted from: pdfAgreement 64%

Explore related topics

Responsible AI Use Scholarly Infrastructure Content Analysis