AI evidence extraction

The 2025 Landscape of Generative AI in Scholarly Writing and Publishing: A Scoping Review of Uses and Ethical Approaches

Lilia Raitskaya, Elena Tikhonova · Journal of language and Education · 2025

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

1.33

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.17323/jle.2025.29876

Methodology & findings

Study design

Scoping review following Arksey and O'Malley (2005) stages and PRISMA-ScR guidelines (Tricco et al., 2018).

Sample

N = 56, 3 groups

Primary method

Mixed-methods approach: (1) Descriptive synthesis of publication characteristics; (2) Within-case analysis of individual studies examining research design, population, AI applications, findings, and research agendas; (3) Cross-case analysis identifying patterns and divergences across corpus; (4) Independent coding by both reviewers with iterative discussion for consensus; (5) VOSviewer keyword co-occurrence analysis (keyword density = 3) for thematic clustering validation; (6) Triangulation of inductively-derived themes with empirical keyword trends. No statistical software specified beyond VOSviewer.

Main result

The analysis of the 56 publications included in this review reveals a significant discrepancy between the formal document type assigned by the Scopus database and the actual methodological nature of the work. A strong cross-disciplinary consensus has crystallized around several foundational principles, most notably that "AI systems cannot qualify as authors and that human researchers must retain ultimate accountability for scholarly work" (Yoo, 2025; Veiga, 2025). The discourse has matured from initial debates about AI authorship to "a sophisticated understanding of how different application patterns, from substantive intellectual tasks to language enhancement, directly shape specific ethical concerns." The thematic analysis reveals "a scholarly discourse that is both complex and structurally coherent, organized around three dominant problem complexes: governance and policy, technological capabilities and risks, and systemic integrity."

Reports effect sizes.

Research paradigm

Mixed-methods synthesis (qualitative thematic analysis + quantitative bibliometric co-occurrence analysis)

Author conclusions

The authors conclude that "This 2025 scoping review has systematically mapped the landscape of literature on generative AI in scholarly writing and publishing, revealing a field characterized by rapid evolution, global engagement, and increasing methodological sophistication." They emphasize that "a strong cross-disciplinary consensus has emerged around foundational principles, particularly human accountability and AI's non-author status, significant challenges persist in balancing transparency with equity, preserving scholarly rigor, and addressing systemic issues like algorithmic bias and 'technological colonialism' in resource distribution." The review provides "a comprehensive foundation for navigating the complex ethical landscape" and "a pathway toward harnessing AI's potential while steadfastly upholding the core values of academic integrity, accountability, and equitable knowledge production that remain fundamental to scholarly endeavour."

Risk of bias

Language bias: Database limited to English-language publications; Database bias: Scopus has coverage gaps in regional publications and certain humanities/social sciences; Grey literature gap: Exclusion of unpublished materials, preprints, and institutional/organizational reports; Misclassification bias: Scopus automated document type classifications were unreliable (10 of 18 reviews miscategorized as articles); Publication bias: Likely overrepresentation of peer-reviewed journal articles; Discipline bias: Selection of Social Sciences, Computer Science, Medicine, Arts and Humanities, Engineering, and Decision Sciences may not capture all relevant discourse; Language bias (English-only inclusion); Database bias (Scopus coverage limitations, gaps in regional and humanities/social science publications); Publication bias (exclusion of grey literature, preprints, unpublished reports); Classification bias (automated Scopus categorization misclassifying review articles as primary research); Geographic bias (potential underrepresentation of non-English speaking regions); Disciplinary bias (coverage gaps in specific humanities and social science disciplines); Selection bias: Limited to English-language publications indexed in Scopus; exclusion of grey literature (policies, preprints, organizational reports); Geographic bias: Coverage gaps in regional publications and non-English databases; Disciplinary bias: Underrepresentation of certain humanities and social science disciplines; Database classification bias: Misclassification of ~53% of review articles as primary research articles in Scopus, requiring manual verification; Publication bias: Reliance on published peer-reviewed literature only; early-stage work and negative findings potentially underrepresented

Limitations

"A primary constraint is its limited inclusion of grey literature, which results in the potential omission of crucial document types such as publisher and institutional policies, preprints from servers like arXiv and SSRN, and formal reports from key organizations like COPE and UNESCO." The authors note "Scopus is known to have coverage gaps in certain regional publications and specific humanities and social science disciplines compared to other databases, which may introduce a geographic or disciplinary bias." Additionally, the review identified significant issues with automated database classification: "nearly half (10 out of 18) of the review articles in our sample misclassified as primary research articles," highlighting "the critical importance of manual, content-based verification in systematic and scoping reviews, as automated database classifications are insufficiently reliable."

Open questions raised

Development and validation of ethical guidelines and frameworks - need for internationally recognised, interdisciplinary ethical standards across different contexts (education, healthcare, non-English academic contexts)
Enhancing detection and integrity tools - improving accuracy of AI-text identification, reducing false positives/negatives, developing tools for detecting AI-generated images, references, and data
Longitudinal and impact studies - lack of understanding of long-term effects on critical thinking, writing skills, authorial voice, and scholarly development; impact on non-native English speakers and underrepresented groups
Human-AI collaboration and user-centred design - understanding cognitive processes in AI-assisted writing, effects of different interaction models on ownership and output quality
Policy development and standardisation - consistency, effectiveness, and global adoption of journal and publisher policies; standardised disclosure mechanisms
Bias, fairness, and decolonial perspectives - investigation of how GenAI perpetuates biases regarding language, culture, and regional representation; application of decolonial lens to global knowledge production

Extracted from: pdfAgreement 59%

Explore related topics

Scholarly Publishing Scholarly Infrastructure Literature Review