AMA Manual of Style and JAMA Network’s Position on AI and Authorship

ICMJE’s generative-AI recommendation is a short, general statement that has been adopted or referenced across thousands of biomedical journals: AI tools cannot be listed as authors, and their use must be disclosed. JAMA and the JAMA Network family of specialty journals follow that same underlying principle, but they don’t stop at the general statement. Their policy is written into the AMA Manual of Style itself, with specific chapter guidance, a required disclosure location, a defined list of elements a disclosure must contain, and separate rules for how peer reviewers and editors may use AI tools when handling a manuscript. This page covers what JAMA Network’s policy actually requires, where it lives in the AMA Manual of Style, and exactly where it goes further than ICMJE’s general position. For the broader multi-publisher landscape, see CASRAI’s guide to journal and publisher policies on generative AI in manuscripts; for the ICMJE baseline itself, see ICMJE Generative AI Policy and ICMJE Authorship Criteria.

Where this policy comes from

JAMA Network’s current position was set out in an editorial by Annette Flanagin, Jacob Kendall-Taylor, and Kirsten Bibbins-Domingo, “Guidance for Authors, Peer Reviewers, and Editors on Use of Artificial Intelligence, Language Models, and Chatbots,” published in JAMA on July 27, 2023 (JAMA. 2023;330(8):702-703). It was issued a few months after ICMJE added its own generative-AI language to Section II.A.4 of the ICMJE Recommendations in the May 2023 update — JAMA’s guidance operationalizes that same principle specifically for JAMA and its network of specialty journals, and its substance has since been folded into the citable reference authors, editors, and peer reviewers are pointed to for correct wording: the AMA Manual of Style, 11th edition, at chapter 5.1.12 (ethical and legal considerations of authorship) and chapter 3.15.13 (acknowledgments, which provides example language for disclosing generative AI use). That’s the practical difference in form between JAMA’s policy and ICMJE’s: ICMJE states a principle in its Recommendations document; JAMA’s version lives in a style manual that JAMA Network’s own author instructions, submission system, and editors all point back to for the exact wording and placement a disclosure needs.

No AI tool can be listed as an author

JAMA Network’s policy is unambiguous on this point: “nonhuman artificial intelligence, language models, machine learning, or similar technologies do not qualify for authorship.” This is the same conclusion ICMJE, Elsevier, Nature Portfolio, Science/AAAS, and essentially every other major publisher have reached, and it rests on the same underlying reasoning — authorship requires the ability to take accountability for the work, approve the final version, and answer for its accuracy and integrity, none of which a software tool can do. See AI as author and AI co-authorship rejection (ICMJE 2023) for the Dictionary-level treatment of why this conclusion follows directly from authorship criteria rather than being an arbitrary rule.

What authors must disclose, and where it goes

Where JAMA Network’s policy becomes more prescriptive than ICMJE’s general statement is in exactly what a disclosure must contain and exactly where it must appear. If an author uses a generative AI tool to create or assist with creating any part of a manuscript’s content — text, tables, or images — the policy requires disclosure in the Acknowledgment or Methods section of the manuscript, including:

• The name of the AI tool, language model, or chatbot used
• Its version number and manufacturer
• The date(s) it was used
• A specific description of how the tool assisted — which portions of the manuscript it contributed to and how
• Confirmation that the author reviewed and takes full responsibility for the accuracy and integrity of that content

Chapter 3.15.13 of the AMA Manual of Style gives authors example acknowledgment language they can adapt for this exact purpose, which is a level of practical, citable detail ICMJE’s general recommendation doesn’t provide on its own. The policy also draws a line that ICMJE’s shorter statement leaves less explicit: basic grammar-checking, spelling, and reference-formatting tools do not require disclosure. The disclosure obligation is specifically about generative or creative use — drafting text, summarizing, generating an image or figure, or otherwise producing content a reader would reasonably want to know originated from a model rather than the author. Submission through JAMA Network’s manuscript system typically includes a direct question asking whether any part of the manuscript’s content was created by generative AI, separating that from routine editing-assistance software. See Generative-AI disclosure statement for how this compares to disclosure requirements at other publishers, and Generative AI for the underlying Dictionary definition.

AI use as part of the research itself, not just manuscript writing

JAMA Network’s guidance also separately addresses a case that’s easy to conflate with manuscript-preparation disclosure but is governed by additional, distinct requirements: when AI or a language model was used as part of the actual study methods — for example, in data analysis, image interpretation, or generating study content that is itself part of the reported research rather than the write-up of it. In that situation, authors are expected to report the AI/language model’s role in the Methods section in the same way any other method would be reported, including details such as the prompts used, relevant bias-management or validation steps, any applicable IRB/ethics approval, and adherence to relevant reporting guidelines for the study type. This is a materially different disclosure obligation from the “I used a chatbot to help draft a paragraph” case above, and conflating the two is a common source of incomplete disclosures.

Peer reviewers and editors: separate rules

JAMA Network’s policy doesn’t stop at authors. It sets out explicit expectations for the two other people who handle a manuscript before publication:

• Peer reviewers may not enter manuscript text, figures, or their own review content into a generative AI tool or chatbot — doing so would disclose confidential, unpublished material to a third-party system, which breaches the confidentiality reviewers agree to when accepting a review assignment. Reviewers must also disclose any use of AI tools in preparing their review that does not itself risk that confidentiality breach.
• Editors do not use AI tools to make individual manuscript decisions (accept, reject, revise). JAMA Network does use AI-adjacent tools for specific administrative and integrity functions — plagiarism detection, metadata parsing, reviewer-matching recommendations, and reference validation — but with human editorial oversight retained over the actual editorial decision.

This three-part structure — separate, specific rules for authors, reviewers, and editors, rather than a single blanket statement — is itself one of the more detailed treatments among major publisher AI policies. Confidentiality-driven restrictions on reviewer AI use specifically are covered in more general terms at AI in peer review.

How this differs from ICMJE’s general position

ICMJE and JAMA Network agree completely on the core conclusion: no AI authorship, and disclosure is required. Where they differ is in specificity and form, not principle:

• Where the rule lives. ICMJE’s position is a short paragraph in its Recommendations document, voluntarily adopted by journals that choose to follow ICMJE. JAMA Network’s version is codified in a specific, numbered style-manual chapter (AMA Manual of Style 5.1.12 and 3.15.13) that JAMA Network’s own author instructions and submission workflow directly reference.
• Required disclosure elements. ICMJE requires disclosure but does not itemize what a disclosure statement must contain. JAMA Network’s policy lists specific required elements: tool name, version, manufacturer, dates of use, a description of the AI’s role, and confirmation of author responsibility.
• Where disclosure must go. ICMJE does not mandate a specific manuscript section. JAMA Network specifies the Acknowledgment or Methods section.
• Scope beyond authors. ICMJE’s generative-AI language is framed primarily around authorship and authors’ use of AI in writing. JAMA Network’s guidance separately and explicitly addresses peer reviewer confidentiality obligations and editors’ own use of AI-adjacent tools — a broader scope covering all three roles in the publication process.
• Research-methods AI use. JAMA Network draws an explicit line between AI used to help write the manuscript and AI used as part of the study’s actual methods, with separate, more detailed reporting expectations (prompts, bias management, reporting-guideline adherence) for the latter. ICMJE’s general recommendation doesn’t spell out this distinction to the same degree.

In practice, this means an author submitting to a JAMA Network journal is not just following “the ICMJE rule plus JAMA’s house style” — they’re following a materially more specific, citable, and role-by-role set of requirements that happens to share ICMJE’s underlying conclusion.

Practical checklist for authors submitting to JAMA Network journals

• Never list an AI tool, chatbot, or language model as an author or co-author.
• If generative AI created or assisted with any manuscript content (text, tables, images), disclose it in the Acknowledgment or Methods section with tool name, version, manufacturer, and dates of use.
• Describe specifically what the AI tool did and confirm you reviewed and take responsibility for that content’s accuracy.
• Answer the manuscript-system’s AI-use question accurately — it is asked separately from routine spelling/grammar tool use, which does not require disclosure.
• If AI was used as part of the study methods themselves (not just the writing), report its role, prompts, and any relevant validation or bias-management steps in the Methods section, and follow the relevant reporting guideline for your study type.
• If you’re a peer reviewer, never paste manuscript content or your review into a chatbot or other generative AI tool.
• Check chapters 5.1.12 and 3.15.13 of the AMA Manual of Style, 11th edition, for the current example acknowledgment wording before drafting your disclosure.

Frequently asked questions

Is JAMA Network’s AI policy different from ICMJE’s?

Not in conclusion — both prohibit AI authorship and require disclosure of AI use. JAMA Network’s version is more specific: it names exactly what a disclosure must contain, where it must appear in the manuscript, and it’s codified in a citable AMA Manual of Style chapter rather than left as a general statement.

Do I need to disclose using a grammar-checking tool like Grammarly?

No. JAMA Network’s disclosure requirement is specifically for generative or content-creating AI use — drafting text, summarizing, or generating images/figures. Basic spelling, grammar, and reference-formatting tools are excluded.

Where in the AMA Manual of Style is this policy written down?

Chapter 5.1.12 covers the ethical and legal considerations of AI and authorship; chapter 3.15.13, in the acknowledgments section, provides example disclosure language authors can adapt.

Can a peer reviewer use ChatGPT to help write their review?

Only in ways that don’t involve entering manuscript content or the review itself into the tool, since doing so would breach the confidentiality obligation reviewers agree to. Any AI assistance that doesn’t create that risk must still be disclosed.

Does this policy apply to every journal published by the American Medical Association?

It applies to JAMA and the JAMA Network family of specialty journals, and is published as JAMA Network’s own editorial guidance and reflected in the AMA Manual of Style, which many other biomedical journals beyond the JAMA Network also use as their house style reference.