Perceptions of Artificial Intelligence in the Editorial and Peer Review Process: A Cross-Sectional Survey of Traditional, Complementary, and Integrative Medicine Journal Editors

A cross-sectional survey of Traditional, Complementary, and Integrative Medicine journal editors reveals that while they recognize the potential of artificial intelligence to support routine editorial tasks, its actual adoption remains limited due to a lack of institutional policies, training, and ethical guidance.

The Gist

Imagine the world of academic publishing as a massive, high-stakes baking competition.

In this competition, scientists (the bakers) submit their recipes (research papers) to a panel of judges (the journal editors). These judges have to taste the dishes, check if the ingredients are fresh, ensure the recipe is original, and decide if it's good enough to be published in the "World's Best Cookbook."

Now, imagine a new, super-smart robot assistant (Artificial Intelligence or AI) has just entered the kitchen. This robot can read recipes instantly, check for spelling mistakes, and even spot if someone copied a recipe from the internet.

This paper is like a survey sent to the head judges of the "Traditional, Complementary, and Integrative Medicine" (TCIM) baking competition. These judges specialize in unique, ancient, and culturally diverse recipes (like herbal remedies, acupuncture, and meditation) that don't always fit the standard "Western" cooking rules.

Here is what the survey found, translated into everyday language:

1. The Judges Know the Robot, But Don't Let It Cook Yet

The Situation: About 70% of the judges know who this robot assistant is and have played with it at home. They've used it to write emails or summarize news.
The Catch: When it comes to their actual job as judges, over 60% have never let the robot help them.

• Analogy: It's like a master chef who owns a high-tech food processor but still insists on chopping onions by hand because they aren't sure if the machine will ruin the flavor of their special, ancient spice blend.

2. What the Judges Would Let the Robot Do

The judges are willing to let the robot handle the boring, repetitive chores. They love the idea of the robot:

• Checking the Grammar: Fixing typos and making sure the recipe is written in clear English (81% support).
• Checking for Plagiarism: Making sure the recipe wasn't stolen from someone else (67% support).
• Analogy: They'd happily let the robot wash the dishes and scrub the pots, but they won't let it decide if the soup tastes good.

3. What the Judges Don't Want the Robot to Do

The judges are very skeptical about letting the robot handle the "human" parts of the job. They don't want the robot:

• Talking to the Bakers: Handling complaints or explaining why a recipe was rejected.
• Making the Final Call: Deciding if a complex, culturally unique recipe is actually "good" science.
• Analogy: They fear the robot is like a robot butler who can't understand that a specific spice is sacred in one culture but offensive in another. The judges worry the robot will miss the "soul" of the recipe.

4. The Big Problem: No Instruction Manual

The biggest issue isn't that the judges hate the robot; it's that nobody taught them how to use it safely.

• The Stats: About 65% of the judges said their journal or publisher has no rules (policies) about using AI. Even worse, most haven't taken any training courses.
• Analogy: Imagine giving a judge a brand-new, complex robot butler but no instruction manual. The judges are saying, "We know this thing is powerful, but we don't know if it's safe to let it near our precious ingredients, and we don't know who to blame if it burns the kitchen down."

5. The Fear of "Hallucinations" and Bias

The judges are worried about two main things:

• The "Lying" Robot: AI sometimes makes things up (called "hallucinations"). If the robot invents a fake ingredient or a fake study, it could ruin the whole cookbook.
• The Biased Robot: If the robot was trained on data that only likes "Western" cooking, it might unfairly reject the "Eastern" or "Traditional" recipes this journal specializes in.
• Analogy: It's like worrying that the robot might accidentally add poison to the soup because it read a fake recipe online, or because it thinks "spicy" is bad just because its training data came from a place that doesn't like spice.

6. The Future: "We Need It, But We Need Rules"

Despite the fears, 83% of the judges believe AI will be very important in the future. They see it as a tool that could save them hours of work.

• The Verdict: They are ready to adopt the robot, but only if someone writes a clear instruction manual (policies) and gives them a cooking class (training) on how to use it without ruining the food.

Summary

The paper concludes that the judges of these unique medical journals are cautiously optimistic. They see the robot as a helpful assistant for washing dishes and checking spelling, but they are terrified of letting it taste the food or make the final decisions.

The takeaway: Before we let AI run the kitchen, we need to teach the judges how to use it, write clear rules about what it can and cannot do, and make sure it doesn't accidentally ruin the special, cultural flavors of the recipes it's reviewing.

Technical Summary

1. Problem Statement

The integration of Artificial Intelligence (AI), specifically AI chatbots (AICs), into scholarly publishing offers potential efficiencies in manuscript screening, plagiarism detection, and workflow automation. However, Traditional, Complementary, and Integrative Medicine (TCIM) presents a unique context for AI adoption. TCIM research involves diverse methodologies, culturally embedded knowledge systems, and interdisciplinary approaches that differ significantly from conventional biomedicine.

Despite the potential for AI to streamline these complex processes, there is a paucity of empirical data regarding how TCIM journal editors perceive, utilize, and regulate AICs. Key gaps include:

• Lack of understanding regarding the specific barriers (practical, ethical, institutional) to AI adoption in TCIM.
• Uncertainty about the alignment of AI tools with the unique methodological and cultural nuances of TCIM research.
• Absence of data on the availability of training and policies guiding responsible AI use within TCIM journals.

2. Methodology

The study employed a cross-sectional, anonymous online survey design.

• Target Population: Editorial board members (Editors-in-Chief, associate editors, board members) of 115 TCIM journals.
• Sampling Strategy:
  • Journals were identified via a structured search of Journal Citation Reports, Scopus, and the US National Library of Medicine (NLM) catalog using specific TCIM subject headings.
  • Editor names and email addresses were manually extracted from journal websites.
  • Exclusion Criteria: Staff involved only in formatting (copyeditors), statistical editors, and non-editorial staff were excluded.
• Data Collection:
  • Instrument: A 38-question structured questionnaire developed based on a previous study by Ng et al. (2025) and validated by TCIM subject matter experts.
  • Content: Covered demographics, familiarity with AI, current usage patterns, perceived benefits/challenges, ethical concerns, and future outlook.
  • Timeline: The survey was open for 6 weeks (May 26 – July 7, 2025) with follow-up emails sent to non-respondents.
  • Analysis: Quantitative data were analyzed using descriptive statistics. Qualitative open-ended responses underwent inductive thematic analysis (coding by two independent authors, consensus building, and thematic grouping).
• Ethical Considerations: The study was approved by the University Hospital Tübingen Research Ethics Board. Data were anonymized, and the study was registered on the Open Science Framework (OSF).

3. Key Contributions

• First Specialized Survey: This is the first comprehensive assessment of AI perceptions specifically within the TCIM editorial landscape, distinguishing it from general biomedical or life science studies.
• Policy Gap Identification: It highlights a critical disconnect between the high potential of AI and the lack of institutional governance (policies and training) in TCIM publishing.
• Nuanced Adoption Profile: The study delineates specific areas where editors are willing to adopt AI (text-focused tasks) versus areas of resistance (interpersonal/interpretive tasks), providing a roadmap for tool developers.
• Thematic Insights: Through qualitative analysis, it uncovers deep-seated concerns regarding "hallucinations," bias, and the erosion of human judgment in culturally sensitive research fields.

4. Key Results

Out of 5,119 invitations, 217 eligible participants completed the survey (5.2% response rate).

Demographics:

• Majority male (54.9%) and female (42.5%).
• Predominantly located in the Americas (40.4%) and Europe (23.3%).
• Most were faculty members (70%) and editorial board members (75.7%).

Familiarity and Usage:

• High Familiarity: 70.0% of editors were familiar with AI tools; 83.7% had used AI chatbots personally.
• Low Professional Adoption: 61.0% had never used AI chatbots for editorial tasks.
• Future Intent: Editors were generally unlikely to use AI in their roles soon (41.1% "unlikely" or "very unlikely").

Perceived Utility (Helpfulness):

• High Support: Strongest support was for language/grammar checks (81.0%) and plagiarism/ethical screening (67.4%).
• Low Support: Significant skepticism regarding interpersonal tasks (e.g., handling author inquiries, communicating status) and complex interpretive tasks.
• Current Usage: The most frequent use reported was for language checks (25.0%) and plagiarism screening (22.1%). Most tasks (71.5% average) were reported as "never" used.

Training and Policy Gaps:

• Lack of Training: 65.3% reported their journal/publisher provided no training on AI tools.
• Lack of Policy: 30.1% confirmed no AI policies existed; 46.0% were unsure.
• Training Needs: 49.1% believed editors need "a lot" of training to use AI effectively.
• Willingness: 63.0% expressed interest in learning more about AI use in publishing.

Perceived Benefits vs. Challenges:

• Benefits: Reducing repetitive administrative burdens (70.1%) and improving manuscript language/clarity (76.0%).
• Challenges:
  • Bias: 82.1% feared perpetuation of biases from training data.
  • Contextual Understanding: 79.3% worried AI lacks the human insight to interpret cultural/emotional nuances.
  • Ethics: 79.1% cited concerns regarding accountability, transparency, and responsible use.
  • Practicality: High agreement on the need for initial time/resource investment (78.5%).

Qualitative Themes:

• "No AI in Authorship/Peer Review": Strong consensus that AI should not make final decisions or be listed as an author.
• "Assist with Peer Review": AI viewed as a tool for discrete tasks (reference verification, grammar) but not for holistic assessment.
• "Privacy and Reliability": Major concerns about data confidentiality (unpublished manuscripts) and the risk of "hallucinations" compromising research integrity.

5. Significance and Implications

• Strategic Guidance for Publishers: The findings suggest that TCIM journals should prioritize targeted, low-risk AI applications (grammar, screening) while delaying high-stakes automation (decision-making, peer review scoring) until robust policies are established.
• Need for Governance: There is an urgent need for TCIM publishers to develop clear, transparent AI policies and mandatory training programs to bridge the gap between editor interest and current capability.
• Cultural Sensitivity: The study underscores that AI tools for TCIM must be culturally calibrated to handle diverse methodologies and non-Western knowledge systems, which current general-purpose LLMs may struggle with.
• Future Research: The results call for longitudinal studies to track how AI adoption evolves as tools improve and policies are implemented, specifically focusing on the preservation of research integrity in TCIM.

In conclusion, while TCIM editors recognize the potential of AI to enhance efficiency, adoption is currently stalled by a lack of institutional support, ethical uncertainty, and the unique complexity of TCIM research. The path forward requires a balanced approach: leveraging AI for administrative support while rigorously maintaining human oversight for critical editorial judgments.