AI in Academic Publishing: Policies, Disclosure, Peer Review, and Researcher Compliance

Key Takeaways:

• Most major journals and publishers do not permit AI authorship. Elsevier, Wiley, Springer Nature, Taylor & Francis, Sage, IEEE, JAMA Network, NEJM Group, The Lancet Group, and AAAS (Science) each require that accountability rest with named human authors.
• Sage, Springer Nature, and Elsevier exempt routine language polishing from disclosure, IEEE recommends disclosing it anyway, and Science expects AI use to be noted in the cover letter as well as inside the manuscript.
• Elsevier, Taylor & Francis, and Springer Nature prohibit peer reviewers from uploading manuscripts into generative AI tools, and The Lancet asks reviewers not to use AI for scientific assessment at all.
• Verification remains the author’s job. Every policy holds named humans responsible for accuracy, references, and images, no matter how much AI assistance was involved.

Mapping AI in Academic Publishing Across 9 Major Publishers

Generative AI tools entered research workflows faster than the rules governing them. The result is a patchwork: broadly similar principles, materially different requirements. This article compares the current published positions of several large publishers and journals.

Who sets the rules, and at what level

Policies operate at 3 levels, and all 3 can apply to a single submission.

• Publisher level: Elsevier, Springer Nature, Wiley, Taylor & Francis, Sage, and IEEE publish group-wide policies that cover hundreds of titles.
• Journal level: JAMA, NEJM, The Lancet, and Science maintain their own editorial policies and instructions for authors.
• Article-type level: JAMA restricts AI drafting for specific formats, so 2 submissions to the same journal can face different rules.

The 3 questions every policy answers

Despite differences in wording, each policy focuses on the following:

• Can AI be an author? Universally no, because authorship requires accountability that only humans can carry.
• What must be disclosed, and where? This is where journals and publishers diverge most, particularly on language editing.
• What is off limits entirely? Data generation, manipulation of research images, and uploading confidential manuscripts into free public AI tools during peer review

Why publisher-level rules are not the last word

Taylor & Francis states plainly that some of its journals may not allow generative AI use beyond language improvement, and advises authors to review the instructions for authors on the journal homepage before submitting.

Elsevier makes a parallel point, noting that the range of AI use varies by journal and directing questions to the specific journal. Sage, meanwhile, reserves editorial discretion to reject a submission at any stage in the publishing process if generative AI was used inappropriately without disclosure. Wiley also cautions authors to make sure that the AI tool they use doesn’t claim ownership of the AI output or interfere with copyright.

 

A side-by-side comparison of journal AI guidelines

 

Publisher or journal	Language editing exempt from disclosure?	Primary disclosure location	Reviewers may upload manuscripts?
Elsevier	Yes for basic grammar, spelling, and punctuation; no for substantive restructuring	Separate AI declaration statement at submission, published with the article	No
Wiley	Yes, if changes are simple grammar and punctuation edits only	AI used for drafting, substantive language editing, or translation must still be disclosed in the Acknowledgments AI used in study design, data collection and analysis, or literature search or synthesis must be disclosed in the Methods section	No
Springer Nature	Yes, for defined AI-assisted copy editing	Introduction or Acknowledgements; Methods for model use	No
Taylor & Francis	No, a specific AI usage disclosure statement is required	Statement within the article	No
Sage	Yes, assistive tools need no disclosure	Methods or Acknowledgements	No, this is listed as inappropriate use
IEEE	Generally outside the policy, though disclosure is recommended	Acknowledgments section	Not addressed in the author guidelines
Science	Editing services should be disclosed in acknowledgments	Cover letter plus methods or acknowledgments	Not addressed in this policy page
NEJM	Disclosure required at submission for AI-assisted technologies	Cover letter and the submitted work	Not addressed in this policy page
The Lancet	Authors are asked to declare AI tool use at submission	Declaration at submission	Reviewers are asked not to use AI for scientific review
JAMA	Governed by the AI sections of the instructions for authors	As directed by the instructions for authors	Not addressed in the instructions for authors

Tools and tasks the journal AI guidelines actually cover

Policies rarely list tools by name. They describe tasks. Map your own workflow against the task categories below.

• Idea generation and exploration: explicitly supported by Taylor & Francis.
• Literature search, synthesis, and classification: supported by Taylor & Francis; Sage requires disclosure for literature review assistance and reference compilation. Wiley recommends using AI as a complementary tool, not a replacement for conventional database searching, and shares a detailed workflow for authors.
• Translation: supported by Taylor & Francis; Sage requires disclosure when translation forms part of the research process.
• Coding assistance: supported by Taylor & Francis and Wiley; Sage requires disclosure when AI writes, enhances, or error-checks code.
• Data visualization support: Sage requires disclosure; Taylor & Francis permits it for research data visualizations. Wiley lists specific image types where AI can or cannot be used at all (for generation or editing).
• Language and readability: exempt at Sage and Springer Nature, exempt only at basic level at Elsevier and Wiley, and requiring a statement at Taylor & Francis.
• Accessibility tools: Elsevier does not require disclosure for specialist disability-related assistive technology used solely for accessibility.

Common misunderstandings by authors reading journal/publisher AI guidelines

• Assuming the publisher policy overrides the journal. Taylor & Francis and Elsevier both defer to individual journal instructions. Wiley gives journal editors final authority to determine whether AI use is appropriate or not in the papers that they handle.
• Treating restructuring as copy editing. Corrections for actual mistakes in grammar (e.g., article usage, subject-verb agreement) or punctuation (e.g., commas, apostrophes, semicolons and colons) look quite “small”. If you see your AI tool restructure sentences or move text around, this is not copyediting but substantive editing, and it’s safer to disclose your use of AI for editing along with how you checked the output.
• Forgetting the cover letter. Science and NEJM both expect AI use to be flagged there, not only inside the manuscript.
• Overlooking article type. JAMA does not permit AI, large language models, or chatbots to draft opinion manuscripts, Letters to the Editor, A Piece of My Mind essays, or poetry.
• Missing version numbers. Taylor & Francis requires the full name of the tool with its version number; Science requires the tool, its version, and the full prompt.
• Ignoring the terms of service. Elsevier requires authors to confirm that a tool does not claim training rights over inputs and does not restrict later publication of outputs.

AI Attribution, Citation, and the Authorship Question

Attribution and authorship are separate concepts, and conflating them is a common error. Crediting a tool for a contribution is required. Listing it as an author is prohibited everywhere.

Why AI attribution never means authorship

Publishers ground the refusal in accountability rather than capability. Authorship carries obligations that a tool cannot discharge.

• Elsevier: each coauthor must be accountable for investigating and resolving questions about accuracy or integrity, must be able to approve the final version, and must agree to its submission.
• Taylor & Francis: authorship requires consenting to publication through a publishing agreement and giving contractual assurances about the integrity of the work, alongside managing copyright and licensing.
• Springer Nature: large language models do not meet its authorship criteria because they cannot take accountability for the work.
• Science: AI-assisted technologies do not meet its authorship criteria and may not be listed as authors or coauthors.
• The Lancet: authorship implies responsibilities and tasks that can only be attributed to and performed by humans.

AI attribution in reference lists and captions

Sage sets out the clearest rule on citation practice. Authors should cite original sources rather than generative AI tools as primary sources within the references. Citing the tool may be appropriate where it created images or visualizations, treated as representative contributions rather than original ones.

Science goes further, stating that sources cited in its journal content may not be authored or coauthored by AI tools. Elsevier is explicit that AI must not be cited as an author.

For images, Elsevier requires disclosure in the caption of each image, naming the specific tool, the version, and how it was used. Springer Nature requires that permitted AI-generated images be clearly labeled as AI-generated, and that non-generative machine learning used to manipulate, combine, or enhance existing images be disclosed in the relevant caption.

AI-assisted code, data visualizations, and translations

• Code: Sage requires disclosure where software or code generated by AI was used to undertake additional research, and where code was enhanced or error-checked with AI assistance.
• Data visualizations: Elsevier requires the model or tool name, the version, and the developer or manufacturer, stated in the Methods section, together with reproducible analytical or statistical methods.
• Translation: Sage requires disclosure where translation of materials formed part of the research process. Taylor & Francis lists translation among its supported responsible uses.
• Third-party attribution: Elsevier requires authors to follow the usage policies of the AI tool itself and ensure proper attribution where the tool requires it.
• Existing artwork: Elsevier requires appropriate attribution and rights-holder permission where images are based on existing artwork or graphics, and prohibits generating likenesses of real people, brands, or identifiable products.

AI Peer Review: Permitted Uses and Hard Limits

What AI peer review assistance publishers still allow

 

Publisher or journal	Permitted reviewer use	Prohibited reviewer use
Elsevier	Improving the language and structure of a review report, and background literature searching, using private tools with disclosure	Uploading a submitted manuscript or any part of it into an AI tool
Wiley	Improving organization, tone, and clarity or translating the comment, also for seeking general information about methodology or statistical analysis without sharing specifics about the manuscript under review	Disclosing any key description of the submitted manuscript to AI
Taylor & Francis	Assisting with improving review language	Generating review reports with AI, and uploading unpublished manuscripts, files, images, or information
Springer Nature	Declaring AI assistance transparently in the review report where a tool supported the evaluation	Uploading manuscripts into generative AI tools
Sage	Using generative AI to help write and format the report, with the reviewer remaining responsible for its content	Uploading or copy-pasting the manuscript into a tool, and generating review reports inappropriately
The Lancet	Not specified	Using AI or AI-assisted technologies to assist in the scientific review of a manuscript

Sage attaches a consequence: reviewers who use generative AI inappropriately to produce review reports will not be invited to review again, and their review will not count toward the final decision.

The confidentiality rule that governs AI peer review

Elsevier explains the reasoning in 2 parts. Uploading a manuscript may violate the authors’ confidentiality and proprietary rights, and where the paper contains personally identifiable information it may breach data privacy rights.

Taylor & Francis frames the same risk as a rightsholder issue, warning that failure to comply may infringe intellectual property. Springer Nature adds that manuscripts may contain sensitive or proprietary information that should not be shared outside the peer review process.

Elsevier’s answer for permitted supportive uses is the private tool test. Read the privacy policy and terms of service for statements on data retention and model training, and prefer paid, enterprise, or locally hosted versions, while verifying rather than assuming that protections apply.

Editors, decision letters, and AI peer review integrity

• Sage prohibits editors from using generative AI to produce decision letters or summaries of unpublished research, and reserves the right to act where confidentiality is breached.
• Elsevier permits editors to use private AI tools only in a supportive capacity, such as improving the language of decision letters, and prohibits uploading manuscripts.
• Taylor & Francis instructs editors to check with their publisher contact before using any generative AI tool unless already authorized.
• Springer Nature requires that editorial and peer review assessments be made and verified by humans.
• NEJM states that all content is reviewed by human editors, who make all editorial decisions about what is published.

Elsevier also discloses its own use of AI on the publishing side: identifying relevant reviewers, matching submission scope, detecting duplicate submissions, running technical and research integrity checks, and supporting proof preparation and copy editing after acceptance.

Where AI in Academic Publishing Is Heading

Publisher-side AI and the human-in-the-loop commitment

Publishers are deploying the same technology they regulate, and are pairing it with explicit human oversight commitments.

• Springer Nature occasionally uses internal tools for accessory content such as key points, editorial summaries, glossary terms, and plain language summaries, always edited and fact-checked, with substantive use declared per article or chapter.
• Elsevier applies AI to reviewer identification, duplicate detection, technical and integrity checks, and post-acceptance production, subject to bias evaluation and data privacy requirements.
• NEJM Group uses AI selectively in editorial workflows and in reader-facing features, while stating that AI supports but never replaces clinical judgment.
• Springer Nature tests its approach against 3 questions: whether it benefits the research community, whether a human is in the loop, and whether it meets principles of fairness, transparency, accountability, and respect.
• Wiley uses AI technologies to match manuscripts with suitable reviewers and also to detect ethical breaches like image manipulation and plagiarism. However, for the latter, a human editor reviews AI output before making any decision about the manuscript. Wiley also indicates that they’re exploring use of AI tools to assist manuscript analysis and improve the quality of peer review comments, but there’s no commitment to these yet.

What to re-check before every submission

These policies carry revision dates for a reason. Elsevier’s journal policy notes an update in June 2026, and Springer Nature states that its AI Ethics and Policy Forums monitor regulatory, technological, and community developments and revise policies accordingly.

• Whether the disclosure threshold for language editing has moved at your target publisher.
• Whether the required placement has changed, particularly the split between cover letter, Methods, and Acknowledgements.
• Whether image rules have loosened or tightened, given that Springer Nature ties its position to unresolved copyright questions.
• Whether the journal has issued instructions that are stricter than the publisher policy.

Frequently Asked Questions

Can AI be used to write a peer review report?

Largely no. Taylor & Francis prohibits using AI tools to generate review reports while permitting AI help with review language. The Lancet asks reviewers not to use AI to assist in the scientific review of a manuscript at all. Elsevier permits only supportive uses such as improving language and structure, using private tools, with disclosure. Sage permits help with writing and formatting the report, but treats using AI to conduct peer review as inappropriate use. Wiley allows AI assistance with organization and tone, but reviewers are not allowed to disclose key information from the manuscript to an AI tool.

What happens if you do not disclose AI use in a journal submission?

Sage reserves the right to reject a submission at any stage of the publishing process where an editor learns that generative AI was used inappropriately without disclosure. For published articles, Sage and the journal editor conduct a joint investigation under COPE guidance and the publisher’s corrections and retractions policy. Taylor & Francis states that submissions replacing core author responsibilities with AI may be subject to editorial investigation.

Can I use AI-generated images or figures in a scientific paper?

Rarely, and never for research data. Springer Nature generally does not permit generative AI images, allowing only narrow exceptions that must be labeled as AI-generated. Science requires explicit editor permission. Elsevier permits AI support for explanatory images such as flow charts and schematics with caption-level disclosure, but bars it from primary research images and graphical abstracts. Taylor & Francis permits conceptual illustrations while prohibiting AI in research and clinical testing outputs.

Do different journals from the same publisher have different AI policies?

Yes, Taylor & Francis notes that some of its journals may not allow generative AI beyond language improvement and directs authors to the instructions for authors on the journal homepage. Elsevier states that the range of AI use varies by journal and invites authors to contact the journal directly. JAMA adds a further layer by restricting AI drafting for specific article types. Wiley allows individuals journals to impose their own restrictions on use of AI and to ask authors to remove or rewrite sections where AI has been used.