Introduction
In September 2025, Elsevier—one of the world's largest and most prestigious academic publishers—released the latest version of its guidelines concerning the use of artificial intelligence in the research and writing processes. Targeting authors, peer reviewers, and editorial board members, this document represents a pivotal moment in the interaction between the scientific community and emerging technologies. Adopting a balanced approach, Elsevier acknowledges the potentials of AI while placing particular emphasis on ethical principles, transparency, and accountability.
This note provides a comprehensive and detailed examination of these guidelines, addressing all aspects of permitted use, restrictions, author and reviewer obligations, and the proper declaration of AI usage.
Elsevier's Core Philosophy: AI as an Assistant, Not an Author
One of the most fundamental principles of Elsevier's guidelines is that AI must be regarded as a supplementary tool and assistant, rather than as the primary author or researcher. This distinction is foundational and underpins all subsequent directives.
Elsevier maintains that AI can prove useful across many stages of research, yet it should never replace human critical thinking, creativity, and accountability. Consequently, the use of AI tools is permitted within clearly defined frameworks, which are elaborated in detail below.
Permitted Uses of Artificial Intelligence
a) Synthesis and Analysis of Complex Texts
One permitted application of AI involves assisting with the synthesis and analysis of intricate scientific literature. Researchers may employ tools such as AI Agents and Deep Research to summarize multiple articles, extract key points, and gain a better understanding of the subject literature. This application proves particularly valuable when researchers face an extensive volume of scientific sources and require rapid yet comprehensive review.
b) Identification of Research Gaps
Through broad analysis of existing literature, AI can assist in identifying underexplored areas, enabling researchers to more rapidly discover innovative and less-investigated topics, thereby steering their research toward greater scientific significance.
c) Ideation and Brainstorming
In the early stages of research, the use of AI to generate preliminary ideas, suggest theoretical frameworks, or offer alternative perspectives is permitted. It should be noted, however, that these ideas must be critically evaluated, refined, and developed by the researcher.
d) Improving Readability and Language Fluency
One of the most common and beneficial applications of AI is enhancing sentence structure, increasing textual fluency, and rectifying linguistic issues. This is especially advantageous for researchers whose native language is not English.
e) Translation and Linguistic Conversion
The use of AI for initial translation or converting text from one language to another is permissible, provided that the author thoroughly reviews, edits, and approves the final translation.
Restrictions and Red Lines
a) Prohibition on Direct Use of AI Output as Manuscript Text
The most significant restriction imposed by Elsevier is that direct AI output must not be used as the primary text of the manuscript. In other words, authors may not simply copy text generated by tools such as ChatGPT, Jasper, or other language models and incorporate it directly into their articles. AI output should serve solely as inspiration, a preliminary draft, or a starting point. The author is obligated to fully revise, analyze, complete, and enrich this output with their own expertise and knowledge.
b) Prohibition of Listing AI as Author or Co-Author
Elsevier explicitly states that AI cannot be credited as an author or co-author of an article. Authorship demands full accountability, critical thinking, and human creativity—qualities that AI lacks. Only natural persons who have actively contributed to the design, execution, analysis, and writing of the research may be recognized as authors.
c) Intellectual Property and Privacy Concerns
Elsevier warns authors that when using AI tools—particularly online platforms—they must ensure familiarity with the terms and conditions of use. Some tools may employ user-input data to train their models, potentially leading to infringement of the researcher's intellectual property rights. Therefore, authors must not enter confidential information, raw research data, or core sections of their manuscripts into public AI platforms unless they are certain that such platforms respect privacy and do not utilize user data for retraining purposes.
The Weighty Responsibilities of Authors: Accountability
A central pillar of Elsevier's guidance is the emphasis on the author's full accountability. Even when AI has been utilized in certain stages of the research process, the author bears complete responsibility for the final content. This accountability encompasses:
a) Verification of Information Accuracy
The author must meticulously review and confirm the accuracy of all AI-generated outputs. A well-known challenge with large language models is the phenomenon of "hallucination," wherein the model produces false information, fabricated references, or spurious data. Elsevier expressly warns that AI-generated references may be entirely fictitious. Consequently, authors are obligated to verify each source individually and ensure its authenticity. Failure to do so may lead to manuscript rejection or even damage to the researcher's academic credibility.
b) Originality and Expression of Personal Ideas
The author must ensure that the final manuscript reflects their own original analysis, interpretation, and ideas. Although AI may assist in early stages, theoretical perspectives, data analysis, conclusions, and scholarly discussion must originate from the author's own reasoning and expertise.
c) Privacy and Data Confidentiality
Authors must not input personal data, sensitive patient information, proprietary corporate data, or any information subject to privacy regulations into AI tools. Such actions not only breach research ethics but may also entail serious legal consequences.
d) Avoidance of Discrimination and Bias
AI may reflect biases present in its training data within its outputs, relating to gender, race, culture, or other factors. The author is responsible for reviewing AI outputs from this perspective and ensuring that the final content contains no unfair discrimination or bias.
Strict Regulations Concerning Images and Figures
One of the most sensitive areas of Elsevier's guidelines pertains to the use of AI in generating, altering, or manipulating scientific images and figures.
a) Absolute Prohibition of Image Generation or Manipulation
Elsevier explicitly declares that the use of AI to generate new images, alter existing ones, remove or add elements to images, or perform any manipulation that changes the scientific content of an image is strictly forbidden. The rationale is clear: the integrity and honesty of scientific data are of paramount importance. Image manipulation can distort results, mislead readers, and ultimately damage scientific credibility.
b) Permissible Adjustments: Basic Image Settings
Elsevier permits only limited adjustments to images, including:
- Brightness adjustment: to improve image clarity
- Contrast adjustment: to better distinguish image elements
- Color balance correction: to rectify improper color representation
It is essential that such adjustments do not obscure, distort, or alter the primary scientific information within the image. All modifications must be applied uniformly across the entire image, rather than to specific portions.
c) Image Forensics Detection Tools
Elsevier warns that it employs advanced Image Forensics tools to detect AI-induced manipulations. These tools are capable of identifying digital alterations, removal or addition of elements, and even AI-generated imagery. Upon detection of manipulation, the manuscript may be rejected, withheld from publication, or—if already published—retracted, which would significantly harm the researcher's scientific reputation.
d) Exception: When AI Constitutes Part of the Methodology
The sole exception arises when the use of AI is itself an integral component of the research methodology—for instance, in medical imaging studies employing AI algorithms for disease diagnosis or image analysis. In such cases, the author must provide a detailed description within the Methods section, including the AI model's name, version, developer, training procedure, parameters used, and limitations. Such information is essential for research reproducibility and accurate result evaluation.
Mandatory Declaration of AI Use
One of the most significant updates in Elsevier's new guidelines is the requirement to provide a formal declaration concerning the use of AI.
a) When Must a Declaration Be Provided?
Whenever an author has utilized generative AI tools such as ChatGPT, GPT-4, Claude, Jasper, AI Agents, or any similar tools during the research or writing process, a formal declaration must be incorporated into the manuscript.
b) Placement of the Declaration
This statement must appear in a separate section, preceding the reference list, under the following heading:
Declaration of Generative AI and AI-assisted technologies in the writing process
c) Exact Content of the Declaration
Elsevier provides a specific template for this declaration:
During the preparation of this work the author(s) used [NAME TOOL / SERVICE] in order to [REASON]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
In this template:
- [NAME TOOL / SERVICE]: the precise name of the tool or service used (e.g., ChatGPT-4, Claude 3, Jasper AI)
- [REASON]: the purpose of using that tool (e.g., to improve text readability, initial translation, or literature summarization)
Practical Example:
During the preparation of this work the author(s) used ChatGPT-4 in order to improve the readability and language quality of the manuscript. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the publication.
d) Importance of the Concluding Sentence
The concluding sentence of the declaration, which emphasizes the author's full responsibility, is critically important. It affirms that the author has reviewed, edited, and approved all final content and accepts complete accountability for it.
Cases Not Requiring Declaration
To avoid excessive bureaucracy and maintain focus on truly significant matters, Elsevier exempts certain tools from the declaration requirement:
a) Grammar and Spell-Checking Tools
Tools such as Grammarly, ProWritingAid, or similar software used solely for grammar, spelling, and punctuation correction do not require declaration. These have long been accepted as standard writing aids.
b) Reference Management Software
Software such as EndNote, Mendeley, Zotero, or RefWorks, used for organizing and managing references, likewise do not require declaration.
c) Statistical and Data Analysis Tools
Standard statistical software such as SPSS, R, Python, MATLAB, and similar tools that have been used in scientific research for years do not need to be mentioned in the AI declaration (though they should be cited in the Methods section).
d) Important Note: Distinction Between Declaration and Methods Section
If AI forms part of the data analysis or research methodology, it should not be mentioned in the end-of-article declaration but rather described fully within the Methods section. The AI declaration pertains exclusively to tools used during the writing and manuscript preparation process, not to tools that constitute the core of the research methodology.
Special Restrictions for Reviewers and Editors
Elsevier harbors serious concerns regarding reviewers' and editors' use of AI and has instituted stringent regulations in this area.
a) Prohibition of Uploading Manuscripts to AI Tools
Reviewers and editors must under no circumstances upload entire manuscripts or portions thereof into public AI tools (such as ChatGPT, Claude, or other platforms). The reasons for this prohibition include:
- Breach of confidentiality: manuscripts under review are confidential and must not be shared with third parties
- Threat to intellectual property rights: uploading manuscript text to platforms that may use it to train their models could infringe upon the author's intellectual property
- Potential information leakage: sensitive data or innovative findings may be inadvertently disclosed
b) Prohibition of Using AI to Write Review Reports
Reviewers must not employ AI to compose review reports or evaluate manuscripts. The rationale is that scientific peer review demands deep critical thinking, expert judgment, and nuanced understanding of the scientific context—capabilities that currently exceed AI's abilities. AI may produce superficial, generic, or even erroneous opinions that neither improve manuscript quality nor facilitate sound acceptance or rejection decisions.
c) Elsevier's Internal AI Tools
Elsevier explains that it employs its own internal, secure AI tools for certain tasks, including:
- Detecting plagiarism and unauthorized text reuse
- Assisting in appropriate reviewer selection
- Identifying potential discrimination or bias within the review process
These tools are designed to maintain confidentiality and ensure that manuscript information is used for no other purposes.
Comparison with Other Major Scientific Publishers
a) Convergence on General Principles
Most major scientific publishers—including Springer Nature, Wiley, Taylor & Francis, PLOS, and others—adhere to the guidelines set forth by the Committee on Publication Ethics (COPE). These guidelines emphasize broadly similar core principles:
- Transparency in AI usage
- Full author accountability
- Prohibition of AI authorship
- Mandatory declaration of generative tool usage
b) Differences in Detail
Despite general convergence, publishers may vary in specific details, including:
- The exact format of the declaration statement
- The level of detail required
- The handling of special cases
Thus, authors are always advised to carefully consult the "Guide for Authors" of their target journal prior to submission.
c) Elsevier as a Gold Standard
Nevertheless, owing to its prominent standing in scientific publishing and its more rigorous standards, Elsevier is often regarded as the gold standard. Adherence to Elsevier's guidelines generally ensures that a manuscript will meet the standards of most other journals as well.
Emerging Challenges and Practical Recommendations
a) Difficulty in Drawing Boundaries
One of the main challenges for authors is precisely delineating the boundary between permissible and impermissible AI use. For instance:
- Using AI to paraphrase a paragraph: Is it permitted? The answer depends on whether the author fully comprehends the content and confirms the paraphrase.
- Using AI to generate an initial draft: Permitted, provided the author thoroughly revises, analyzes, and enriches the draft with their own knowledge.
- Using AI to generate conclusions: This is risky, as conclusions must directly stem from the research findings and the author's analysis.
b) Practical Recommendations for Authors
To ensure full compliance with Elsevier's guidelines, authors may consider the following:
- Stepwise and mindful usage: employ AI only at stages where you fully understand its workings. Never blindly trust its outputs.
- Detailed documentation: maintain notes on how AI was utilized—this aids in drafting an accurate declaration and enables clear responses to editor or reviewer inquiries.
- Multi-step reference verification: personally verify every source obtained through AI. Consult the original source and confirm its existence.
- Use reputable tools: whenever possible, employ reputable professional AI tools with transparent privacy policies that do not use user data for retraining.
- Consult colleagues: discuss AI usage with colleagues and advisors. Diverse perspectives can facilitate a better understanding of ethical boundaries.
c) Consequences of Guideline Violations
Violations of Elsevier's guidelines can entail serious repercussions:
- Desk rejection: if editors or reviewers detect inappropriate AI use, the manuscript may be rejected without further review.
- Retraction of published articles: if violations are discovered post-publication, the article may be formally retracted, leaving a lasting stain on the researcher's academic record.
- Damage to scientific reputation: scholarly reputation built over years can be swiftly undermined.
- Submission bans: in severe cases, the researcher may be prohibited from submitting to Elsevier journals for a specified period.
- Academic and professional consequences: research ethics violations can carry serious implications within academic or professional settings.
The Future of AI in Scientific Publishing
a) Continuous Evolution of Guidelines
The current Elsevier guidelines are not an endpoint but rather the beginning of an ongoing dynamic process. As AI technologies evolve, so too must the guidelines. Elsevier has committed to regularly reviewing and updating its policies.
b) More Advanced Detection Tools
It is anticipated that tools for detecting AI usage in scientific texts will become increasingly sophisticated. These tools may not only identify AI-generated content but might also assess the extent of human involvement in the content.
c) Integration of AI into the Publication Process
In the future, publishers themselves will likely provide authors with authorized, controlled AI tools that comply with ethical standards and maintain confidentiality.
d) Training and Culture Building
One of the most pressing current needs is educating researchers—especially graduate students—on the ethical and responsible use of AI. Universities and research institutions should offer training courses and workshops in this area.
Case Studies: Real-World Scenarios
To better illustrate the application of Elsevier's guidelines, we examine several realistic scenarios:
Scenario 1: Non-native English-Speaking Researcher
Situation: Dr. Ahmadi, a researcher whose native language is Persian, writes his manuscript in Persian and subsequently uses ChatGPT to translate it into English. He then employs the same tool to improve sentence structure and enhance fluency.
Analysis: This usage is permissible, provided that:
- Dr. Ahmadi carefully reviews the translation and ensures the original meaning is preserved
- The scientific content, ideas, and analysis are entirely his own
- He includes a declaration stating that ChatGPT was used for translation and readability enhancement
Appropriate Declaration:
During the preparation of this work the authors used ChatGPT-4 in order to translate the manuscript from Persian to English and improve the readability and language quality. After using this tool, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.
Scenario 2: Using AI for Literature Review
Situation: Dr. Mohammadi wishes to conduct a comprehensive review of 200 recent articles in her field. She provides titles and abstracts to an AI Agent, requesting identification of major themes, research gaps, and overarching trends.
Analysis: This usage is permissible; however:
- Dr. Mohammadi should not directly copy the AI output into her literature review section
- She must read the key articles herself and provide her own analysis
- She may use AI-generated findings as a guide to focus on specific topics
- She should declare the use of an AI Agent for preliminary literature organization
Scenario 3: Generating a Diagram with AI
Situation: Dr. Rezaei wishes to create a conceptual diagram to illustrate his theoretical framework and employs an AI tool to generate it.
Analysis: This is a complex issue:
- If the diagram is purely illustrative or symbolic without scientific data, it may be acceptable, though using standard design tools is preferable
- If the diagram contains scientific data or results, AI generation is prohibited
- The best approach is to use standard scientific software such as R, Python (matplotlib), GraphPad, or Origin for diagram production
Scenario 4: Statistical Analysis with AI Assistance
Situation: Dr. Karimi possesses complex data and uses an AI model to select appropriate statistical tests and interpret results.
Analysis: This practice is hazardous:
- Selecting an appropriate statistical test requires deep understanding of data nature and statistical assumptions
- AI may offer inappropriate recommendations
- If Dr. Karimi lacks sufficient statistical expertise, he should consult a professional statistician rather than AI
- Using AI to interpret statistical results is highly risky and may lead to erroneous conclusions
Recommendation: In such cases, collaboration with a qualified statistician who may be recognized as a research collaborator is far more appropriate.
Scenario 5: Peer Reviewer
Situation: Dr. Nouri receives a manuscript for review and does not fully comprehend certain technical sections. He considers using ChatGPT to explain them.
Analysis: This practice is strictly prohibited:
- Dr. Nouri must not place any part of the manuscript into public AI tools
- If he lacks the necessary expertise to evaluate the manuscript, he should inform the editor and decline the review invitation
- If clarification is required, he should consult expert colleagues rather than AI
The Role of Universities and Research Institutions
a) Establishing Institutional Policies
Universities and research centers should develop clear policies regarding AI use in research that:
- Align with scientific publisher guidelines
- Are transparent and comprehensible to faculty and students
- Are regularly updated
b) Training and Empowerment
Research institutions should:
- Offer training courses on the ethical use of AI
- Provide practical workshops on tools and correct usage
- Develop written and video educational resources
c) Creating Secure Infrastructure
Universities may:
- Provide reliable, secure AI tools to their members
- Establish agreements with AI providers that guarantee privacy
- Launch internal platforms for safe AI utilization
d) Establishing Ethics Committees
Research ethics committees should:
- Update their guidelines to encompass AI-related issues
- Provide consultation in ambiguous cases
- Investigate and pursue violations when necessary
Conclusion and Final Recommendations
Elsevier's September 2025 update demonstrates that the scientific publishing world is moving toward a realistic and responsible acceptance of AI. This approach is founded not on fear and prohibition, but on maximum transparency and accountability.
Key Principles for Success:
- Full Transparency: Always be honest about AI usage. Concealing this information can carry far more serious consequences than disclosing it.
- Accountability: Remember that ultimate responsibility for the final content rests with you, not AI. Never blindly trust generated outputs.
- Mindful Usage: Use AI as a powerful assistant, not as a replacement for your own thinking and analysis.
- Continuous Learning: Guidelines are constantly evolving. Always consult the latest publisher guidelines.
- Consultation and Collaboration: When in doubt, consult colleagues, advisors, or journal editors.
- Ethics Precede Technology: Research ethical principles always take precedence over the use of new technologies.
Ultimately, AI is a tool that can facilitate scientific research, but it cannot—and should not—replace human talent, creativity, and integrity. Researchers who understand and observe this balance will both benefit from the advantages of this technology and safeguard their scholarly reputation.
By adhering to these guidelines, we can continue to produce high-quality, transparent, and trustworthy science in the age of artificial intelligence, thereby contributing to the advancement of human knowledge.