AI and Authorship: Systematic Review with ☸️SAIMSARA

• Generated: 2025-10-12 23:18:37 CEST
• Plan: Premium (expanded craft tokens; source: Europe PMC)
• Source: Europe PMC
• Scope: All fields
• Keyword Gate: Fuzzy (≥60% of required terms, minimum 2 terms matched in title/abstract)
• Total Abstracts/Papers: 46333
• Downloaded Abstracts/Papers: 46333
• Included original Abstracts/Papers: 329
• Total study participants (naïve ΣN): 1731340

• Detection Challenges: Both human evaluators and AI detection tools demonstrate limitations in accurately identifying AI-generated content, particularly in hybrid or longer texts, with human accuracy ranging from 19.6% [85] to 79.41% [161], and AI tool accuracy from 42.9% [190] to 100% [155].
• Perception and Credibility: Content attributed to AI often experiences devaluation, reduced authenticity, and lower perceived credibility (e.g., perceived message credibility reduced by d = 0.36 [118], positive word of mouth reduced [3]), despite AI-generated text sometimes scoring higher in readability or quality [7, 30, 76, 87, 92, 134, 185].
• Ethical and Policy Frameworks: A strong consensus prohibits AI from being listed as an author, emphasizing human accountability, disclosure, and transparency (e.g., 94.4% of Korean medical journals prohibited AI authorship [90], 72% of vascular surgery journals had GenAI policies prohibiting AI authorship [12]), yet implementation and standardization remain inconsistent [1, 13, 20, 33, 139].
• AI as a Co-creator and its Impact on Creativity: AI functions as a textual co-creator or cognitive partner, facilitating ideation and manuscript preparation, but raises concerns about originality, homogenization of voice, and intellectual property [10, 21, 32, 37, 53, 79, 91].
• Quality and Efficiency of AI-Generated Content: AI can improve readability (e.g., ChatGPT-generated personal statements scored higher in readability [30], AI-generated articles had higher readability scores [76]), clarity, and efficiency in writing tasks, sometimes outperforming human-authored content in specific metrics, but may lack depth, technical accuracy, or empathy [9, 31, 52, 107, 150].
• Disparities in Authorship: Gender and geographical disparities persist in AI-related research authorship, with underrepresentation of women (e.g., female first authors in urology-related AI research were 17.6% [48], in AI fairness in healthcare literature white, male authors from high-income countries were overrepresented [169]) and authors from low- and middle-income countries [1, 187, 275, 290].
• Academic Integrity and Misinformation Risks: AI tools pose threats to research integrity through pseudoscientific content, hallucinations, plagiarism, and the potential for "infodemics" and fraudulent authorship (e.g., LLMs threaten research integrity through pseudoscientific content [1], 48 out of 53 examined articles appeared to be AI-generated with fraudulent authorship [56]).

• Medical Education: The inability of educators to reliably distinguish AI-generated from student-written work [38, 55, 151] necessitates new pedagogical approaches and guidelines for academic integrity in medical training.
• Patient Information: While AI can improve readability of patient education materials [158], the potential for misinformation and reduced patient satisfaction when AI involvement is disclosed [19, 122] requires careful oversight and transparency in healthcare communication.
• Research Integrity: The high rates of fraudulent authorship [56] and fabricated citations [215] enabled by AI tools threaten evidence-based medicine [27] and demand stringent editorial policies and human accountability in medical publishing [1, 174].
• Clinical Documentation: Hybrid human-AI collaboration in operative notes improves efficiency and consistency [31], suggesting AI can enhance clinical workflows if human oversight ensures accuracy and depth.
• Psychological Advice: AI-generated psychological advice can be rated favorably for empathy and quality [15], but the inability of recipients to distinguish AI from expert human authorship [6, 15] underscores the need for clear disclosure in sensitive contexts.

• Robust Detection Benchmarks: Future research should establish standardized benchmarks for AI content detection tools, evaluating their reliability across diverse text types, lengths, and hybrid human-AI compositions to inform academic misconduct decisions [24, 36].
• Perceptual Bias Mitigation: Studies are needed to explore interventions that mitigate negative biases against AI-authored content while maintaining transparency, especially in creative and emotional domains where authenticity is highly valued [3, 46, 47, 112].
• Standardized Policy Efficacy: Research should evaluate the effectiveness of current AI authorship policies across various disciplines and publishing venues, identifying best practices for disclosure, accountability, and preventing fraudulent authorship [14, 20, 139].
• Co-Creative Models' Impact: Investigations into the long-term effects of AI as a co-creator on human originality, critical thinking, and the potential for homogenization of artistic and scientific voices are warranted [17, 23, 37, 79].
• Authorship Disparity Interventions: Studies should develop and test interventions to address gender and geographical disparities in AI-related research authorship, promoting equitable participation and recognition [48, 169, 275, 290].

• Heterogeneous Study Designs — The variability in study designs, from unspecified to RCTs, limits the ability to draw strong causal inferences or conduct robust quantitative synthesis.
• Inconsistent AI Tooling — The rapid evolution of AI models and detection tools means findings may quickly become outdated, and comparisons across studies using different AI versions are challenging.
• Subjective Outcome Measures — Many studies rely on subjective perceptions of quality, credibility, or authenticity, which can vary widely among participants and contexts.
• Limited Generalizability — Studies often focus on specific populations (e.g., medical students, consumers) or content types (e.g., essays, abstracts), restricting the generalizability of findings to broader contexts.
• Lack of Long-Term Data — Most studies are cross-sectional or short-term, precluding insights into the long-term impacts of AI on authorship, research integrity, or human-AI collaboration dynamics.

• Develop AI Detection Standards — Establish robust, continuously updated benchmarks for AI content detection across diverse text types and hybrid compositions.
• Explore Human-AI Co-creation — Investigate optimal models for human-AI co-creation that enhance originality and mitigate homogenization of voice.
• Standardize Authorship Policies — Implement and rigorously evaluate standardized AI authorship and disclosure policies across academic institutions and publishers.
• Assess AI's Ethical Impact — Conduct longitudinal studies on the ethical implications of AI in research, including bias, plagiarism, and misinformation.
• Promote Equitable AI Authorship — Develop and test interventions to address gender and geographical disparities in AI-related research authorship.