Report the role of the tool and check the evidence
How to Cite ChatGPT in AMA and Disclose AI Use
Separate AI-assisted editing from research methods, keep model outputs distinct from supporting literature, and write a disclosure that matches what happened.
To decide how to cite ChatGPT in AMA, start with what the tool did. Did the tool revise your wording, take part in your research, or produce the response you are analyzing? Each use needs a different explanation. We checked JAMA’s author instructions, ICMJE recommendations, and UNE’s AMA guide on October 31, 2026. Their guidance helps separate disclosure of the tool’s role from the literature supporting your claims.
Identify how AI was used
Start with the action you performed. If you asked for clearer wording in a paragraph you had already written, record that editing task and the affected section. If a model classified material or produced measurements used in your analysis, record it as part of the research procedure. When the response itself is what you are examining, preserve the particular output you analyzed.
Keep these roles separate in your notes. A study can use AI for classification and later for language editing. Reporting only the editing step would leave the classification procedure unexplained. Conversely, a tool used solely to polish sentences should not be described as having conducted the study.
ICMJE excludes chatbots from authorship and holds human authors responsible for submitted material. It also rejects AI-generated material as a primary source of evidence. A response can therefore be the object of your investigation while the research literature remains the evidence supporting your scientific claims.
Check the journal or course disclosure policy
OUP’s public update notice records the addition of section 3.15.13 on LLMs and chatbots in August 2024. That notice confirms an update; it does not reproduce the detailed section. UNE’s public AMA guide generally places AI references in the text rather than the reference list and calls for a study disclosure describing the tool, prompts, and incorporation of its output.
Use your destination’s current instructions to settle placement. ICMJE recommends explaining AI use in the cover letter and the appropriate part of the submitted work. JAMA places manuscript assistance in Acknowledgments and research use in Methods. Its basic grammar and spelling tools are exempt from the manuscript-assistance guidance. A generative rewriting task needs to be assessed against the actual policy, even when its purpose is improving English.
JAMA also prohibits AI drafting of Opinion manuscripts, Letters to the Editor, and Online Comments. Disclosure does not make a prohibited activity acceptable. For coursework, check the assignment’s permitted uses before borrowing wording from a journal policy.
Draft an AMA AI use statement for language editing
The following original draft describes a permitted editing task. Replace each bracketed field with your records and retain only statements that are true. The placement and field checks use JAMA as the named destination; another journal or course may ask for different details.
Adaptable editing disclosure draft for Acknowledgments
The authors used [tool name], [version and extension identifiers as recorded], from [manufacturer], on [dates of use] to revise the wording of [specific manuscript sections]. The tool was used for language editing only. The authors reviewed the changes against their original text and sources, made the final editorial decisions, and take responsibility for the integrity of the submitted content.
If the version was not recorded or displayed, explain that limitation to the editor instead of guessing a model identifier.
Draft a Methods disclosure for research use
The next original draft illustrates a hypothetical text-classification procedure. Supply the actual protocol and evaluation findings before using it to describe a study. Its brackets identify information you need to provide, including where a reader can find the prompts and results.
Adaptable research-use draft for Methods
We used [platform and tool], [recorded version], from [manufacturer], on [dates] to assign [defined labels] to [specified study material]. The prompt sequence and revisions are provided in [protocol or supplement location]. We evaluated outputs against [specified reference standard] using [prespecified assessment procedure] and handled disagreements through [documented review process]. [Results location] reports the evaluation findings and observed errors. The authors reviewed the outputs and take responsibility for the reported analysis.
Complete the study design, input selection, evaluation, and any applicable ethics, rights, or data-sharing details required by the target journal.
Check each draft against its target policy
JAMA’s author instructions supply the requirements in the first column. A brief editing statement can name the affected prose. Research use requires enough additional detail to explain and evaluate the procedure. Neither example can supply facts the author has not recorded. The reference standard and disagreement review are proposed choices for this hypothetical classifier; select methods suited to your study.
| Policy check | Editing draft | Research-use draft |
|---|---|---|
| Placement | Acknowledgments | Methods |
| Tool identity | Name, version and extensions, manufacturer | Platform/tool, version, manufacturer |
| Timing and role | Dates and affected manuscript sections | Dates and specific research task |
| Prompt reporting for LLM research | Editing role described | Prompt sequence and revisions in supplied protocol |
| Review and responsibility | Authors review changes and accept responsibility | Authors review outputs and accept responsibility |
| Research evaluation | No research procedure claimed | Reference standard, assessment, errors and results need real documentation |
Record the output and verify the original evidence
Keep a working log as you use the tool. Save its displayed identity, the dates, and your instructions. Preserve the response and your subsequent changes where your institution permits this. Distinguish an interface label from an underlying version you cannot establish. If a version was unavailable, record that limitation while you can still check it.
When the response is your object of analysis, identify the saved output precisely enough to locate the passage being discussed. Record the prompt that produced it and whether later prompts changed the response. Select excerpts that your readers can connect to the documented procedure. Follow the journal’s directions for any transcript or supplement, including access restrictions and permissions.
For literature suggested by a chatbot, open the publisher record and read the relevant source before using it. Compare the title and authors, then check that the DOI resolves to the same work. Read the passage supporting your claim. A real paper with a similar title can still be the wrong source, and a correct DOI does not establish that your interpretation is sound.
JAMA advises against using AI or chatbots to generate or format references and permits standard reference managers instead. An ordinary reference formatter does not automatically perform the same task as a generative chatbot. Check the actual workflow and the destination policy before deciding what belongs in an AMA AI disclosure.
Finish the reference list from checked records
Keep numbered literature references attached to the sources you actually used. UNE’s general in-text approach to AI helps explain why an AMA ChatGPT citation need not become another numbered entry. Use the model description to identify the tool and the literature references to support scientific claims.
Before submission, compare the manuscript’s disclosure with the cover letter and any required submission fields. Check that the dates agree and that a coauthor’s additional AI use has been included. Remove any assertion of validation or human review that the team cannot document.