JAMA's AI Policy: What You Can Use, What You Must Disclose, and Where the Lines Are

JAMA expects more detail in the AI disclosure than most journals. Here's what a properly formatted statement looks like:

"During preparation of this manuscript, the authors used ChatGPT (GPT-4, OpenAI, San Francisco, CA) to assist with editing the Discussion section for clarity and conciseness. The authors reviewed and revised all AI-generated suggestions and take full responsibility for the content of the published article."

Key elements JAMA wants:

• Tool name (ChatGPT, Claude, Copilot, etc.)
• Version (GPT-4, Claude 3.5, etc.)
• Manufacturer and location (OpenAI, San Francisco, CA)
• Specific use case (editing the Discussion, improving language, generating code for statistical analysis)
• Author responsibility statement (confirming human oversight and accountability)

A statement that wouldn't pass JAMA's standards:

"AI tools were used during manuscript preparation."

This is too vague. It doesn't tell the reader what tool was used, what it did, or which parts of the manuscript it touched. JAMA's editors will ask you to revise.

Multiple AI tools

If you used more than one tool, disclose each separately:

"The authors used ChatGPT (GPT-4, OpenAI) to improve the readability of the Introduction and Claude (Claude 3.5, Anthropic, San Francisco, CA) to assist with drafting the statistical analysis code in R. All AI outputs were reviewed and verified by the authors, who take full responsibility for the published content."

JAMA's threshold for disclosure is lower than many researchers expect. Here's the breakdown:

The dividing line is roughly this: if the tool uses generative AI or large language models, disclose it. If it's a traditional software tool (spell checkers, reference managers, statistical packages), you don't need to.

JAMA treats undisclosed AI use as a publication ethics violation, similar in severity to undisclosed conflicts of interest or undisclosed duplicate submission. The consequences escalate based on when the violation is discovered:

During peer review: The manuscript is returned to the authors with a request to add proper AI disclosure. If the omission appears deliberate, the editor may reject the paper outright. JAMA editors have noted in editorials that transparency failures undermine trust in the entire manuscript.

After acceptance but before publication: The production team adds the disclosure to the Methods section with the authors' consent. If authors refuse to disclose, the acceptance may be rescinded.

After publication: This is where consequences get serious. JAMA can issue:

• A correction (if the AI use was minor and didn't affect scientific content)
• An expression of concern (if the scope of undisclosed AI use is unclear)
• A retraction (if AI-generated content is found to contain fabricated data or unverifiable claims)

JAMA also follows ICMJE guidelines on publication ethics investigations. A finding of deliberate non-disclosure could result in the authors being flagged to their institution and to other journals where they serve as authors or reviewers.

The career risk is real. JAMA publishes roughly 11% of submitted manuscripts. Getting your paper through that filter and then having it corrected or retracted for an avoidable disclosure issue would be a significant setback. Over-disclosure costs you nothing. Under-disclosure can cost you the paper.

JAMA's AI policy applies uniformly across these titles:

JAMA Network Open deserves special mention. It's the network's only fully open-access journal, with a much higher acceptance rate than JAMA itself (~20% vs ~11%). The AI policy is identical, but the volume of submissions is significantly higher, meaning more AI-assisted manuscripts flow through its editorial system.

JAMA moved fast. The editorial acknowledging AI's impact on medical publishing appeared within weeks of ChatGPT's widespread adoption. The journal didn't wait for ICMJE consensus, it set its own rules first and aligned with ICMJE updates as they came.

Before you write:

• Decide upfront which AI tools you'll use and for what purpose. This makes the disclosure statement easier to write later.
• Keep a log of AI interactions. If you use ChatGPT to edit three paragraphs, note which paragraphs and what prompts you used. You probably won't need this level of detail for the disclosure, but having it protects you if questions arise.

During manuscript preparation:

• Don't use AI to generate clinical data, patient descriptions, or case details. JAMA's editors are experienced enough to spot fabricated clinical scenarios, and AI-generated medical content carries serious accuracy risks.
• If you use AI for statistical code, verify every output independently. JAMA's statistical reviewers will check your methods regardless.
• Draft your disclosure statement as you write, not after. It's easier to remember what you used AI for while you're doing it.

At submission:

• Include the AI disclosure in the Methods section, formatted as described above.
• If you didn't use any AI tools, you don't need a negative disclosure statement. JAMA doesn't require authors to state "no AI tools were used."
• Double-check that all co-authors are aware of the AI disclosure. JAMA requires all authors to approve the final manuscript, and that includes the AI disclosure.

Common mistakes:

• Listing ChatGPT in the Acknowledgments section instead of Methods. JAMA specifically requires Methods placement.
• Disclosing AI use for the abstract but not specifying which sections of the main text were AI-assisted.
• Forgetting to include the version number. "ChatGPT" isn't specific enough, JAMA wants "ChatGPT (GPT-4)" or "ChatGPT (GPT-3.5)."
• Assuming your institution's AI policy overrides JAMA's. If your university allows unrestricted AI use, that doesn't mean JAMA will accept an undisclosed AI-written manuscript.

It's worth noting that JAMA has been transparent about exploring AI tools in its own operations. The journal has published research on AI's ability to answer medical exam questions, generate clinical text, and detect errors in manuscripts. JAMA's editors aren't anti-AI, they're pro-transparency.

This matters because some researchers worry that disclosing AI use will bias editors against their manuscript. JAMA's published position suggests the opposite: editors are more concerned about hidden AI use than disclosed use. A paper with a clear, honest AI disclosure is better positioned than one where undisclosed AI use is later discovered.

A pre-submission manuscript review can catch statistical reporting gaps and scope mismatches before JAMA's 3-day desk triage.