AI evidence extraction

Ethical Use of Artificial Intelligence in Academic Journal Writing: A Systematic Review Analysis

Luqman Affandi, Didik Dwi Prasetya, Hakkun Elmunsyah · International journal of research and scientific innovation · 2025

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

10/10

Relevance

1/4

Quality (LMQS)

Evidence

Citations

0.44

FWCI

Top 10%

Impact

This is an AI analysis. Read the peer-reviewed original at the publisher: https://doi.org/10.51244/ijrsi.2025.12120001

Methodology & findings

Study design

Systematic Literature Review (SLR) following PRISMA 2020 guidelines.

Sample

N = 44, 4 groups

Primary method

Quantitative elements: Descriptive analysis of prevalence of AI policies, distribution of guidelines across publishers, and bibliometric patterns. Qualitative elements: Thematic analysis to identify recurring concerns, inconsistencies in policy implementation, ethical debates, and emerging best practices. PICCO framework used to structure literature scope.

Main result

The study found that "96.7% of the analyzed literature expresses ethical concerns related to AI use, including plagiarism, lack of originality, risks of inaccurate content, and the phenomenon of hallucination." Additionally, "only 20-30% of publishers have comprehensive AI policies, while 30-50% lack any formal guidance, contributing to regulatory fragmentation and uncertainty for researchers," and "despite claimed detection accuracies of 94-99%, real-world performance averages only ~26%, undermining confidence in detection-based enforcement."

Reports effect sizes and confidence intervals.

Research paradigm

Interpretivist/Qualitative with descriptive quantitative synthesis

Author conclusions

The authors conclude that "Ethical concerns overwhelmingly dominate the literature, with 96.7% of studies reporting risks related to plagiarism, loss of originality, authorship ambiguity, and AI hallucinations." They further state that "purely technical or policy-driven solutions are inadequate. Instead, the literature converges on the need for an integrated, principle-based approach combining mandatory transparency and disclosure of AI use, sustained ethics education and AI literacy, adaptive discipline-specific frameworks, and meaningful human oversight to ensure that generative AI ultimately strengthens rather than undermines academic integrity and trust in scholarly knowledge."

Risk of bias

Publication bias: Only peer-reviewed articles from Springer database were included, potentially excluding relevant grey literature; Language bias: Only English-language publications were included; Temporal bias: Limited to 2021-2025 publications, which may not capture earlier foundational work; Selection bias: Article selection based on topic relevance screening, with 312 non-relevant items excluded, reducing sample to 44 studies (8.8% of initially identified records); Database selection bias: Limited to Springer publications only, may not capture full landscape of AI ethics research from other major databases; Language bias: English-language publications only; Publication bias: Peer-reviewed articles only; grey literature excluded; Selection bias: Study selection conducted by review team without reported inter-rater reliability metrics; Heterogeneity of included studies: Mix of empirical research, literature reviews, and editorial pieces with varying methodological rigor; Temporal bias: Publication date range 2021-2025 may miss foundational earlier work, though justified by rapid AI development; Selection bias: Search limited to Springer database only; may miss relevant literature from other databases; Publication bias: Only peer-reviewed English-language articles included; grey literature excluded; Timing bias: Focused on 2021-2025 period; may underrepresent earlier foundational work; Outcome reporting bias: Study extraction focused on authors' reported findings without independent verification; Subjective thematic analysis: Qualitative coding conducted without reported inter-rater reliability measures

Limitations

The authors acknowledge several critical limitations: "At present, no single method is capable of consistently detecting all forms of AI-generated content with high accuracy," and "reliance solely on technical detection approaches is insufficient for safeguarding academic integrity." Additionally, the review notes that "approximately 22% of proposed recommendations remain largely principle-oriented and difficult to operationalize into concrete technical or procedural guidelines" and "60.9% of detection research relies on fragmented and non-standardized methods." The review is limited to Springer database publications only and to English-language sources, potentially missing relevant studies from other databases and non-English literature.

Open questions raised

Robust and Scalable Verification Methods: No gold-standard, scalable approach currently exists to reliably verify AI use across diverse publication types
Motivational and Psychological Factors: Limited understanding of drivers underlying ethical and unethical AI use
Discipline-Specific Guidelines: Most existing guidelines remain generic; need for discipline-tailored frameworks
Longitudinal Impact Assessment: Scarcity of long-term studies evaluating sustainability and effectiveness of AI-related policy interventions
Global Harmonization Frameworks: Absence of universally adopted international framework to align ethical AI policies
Faculty AI Literacy: Substantial gaps in faculty understanding of AI capabilities, limitations, and ethical implications

Extracted from: pdfAgreement 74%

Explore related topics

Responsible AI Use Research Integrity