Using AI responsibly
“How do I write an AI-use disclosure my journal accepts?”
Copy-and-adapt disclosure sentences, matched to what you actually did.
You used an AI tool somewhere in the work, and now a form or a reviewer wants a disclosure — but you have seen articles warning against AI use and you are not sure what is safe to write. The honest answer is a short, specific sentence. Here are four you can adapt.
“How do I use GPT Chat to write my dissertation effectively without any problems? I found a number of articles advising against using it…” — a student, unsure what is acceptable
The problemYou know what you did. You don’t know how to say it.
Disclosure anxiety is rarely about the AI use itself — brainstorming with a model, or asking it to tighten a paragraph, is now routine and permitted by most journals. It is about the wording. Say too little (“AI was used”) and it reads as evasive; say it wrong and you worry it sounds like the AI wrote the paper. The fix is to describe the specific task in one plain sentence, matched to what actually happened.
Why it mattersA vague disclosure is worse than a precise one.
“AI was used in the preparation of this manuscript” tells a reader nothing and invites suspicion of everything. A precise statement — which tool, which version, which task, and that a human reviewed the output — closes the question instead of opening it. It also protects you: if the disclosure matches your records, there is nothing to unravel later.
Common mistakesFour disclosures that cause trouble
- “AI was used.” No tool, no task, no version. It is not a disclosure, it is a shrug — and reviewers read it as one.
- Over-disclosing trivial edits. A grammar checker usually does not need a paragraph. Know your venue’s threshold; do not manufacture a problem.
- Listing AI as an author. Every major guideline says a tool cannot be an author, because it cannot take responsibility. Disclose its use; do not credit it.
- A disclosure that does not match your records. If you write “editing only” but the tool drafted a section, the mismatch is the risk — not the AI.
The minimal explanationWhat counts as AI use, and what to name
Disclose when a generative tool contributed to the content or analysis — brainstorming, drafting, summarising sources, editing beyond spell-check, or any step in your data analysis. Policies vary, so check the target venue, but a good statement always names four things: the tool and version, the task it did, the stage at which you used it, and the human review that followed. The last is the load-bearing part: it records that a person, not the model, made the decisions.
Worked exampleFour statements to adapt
Match the one to what you actually did
Brainstorming. “During the early planning of this study, ChatGPT (GPT-4o, OpenAI) was used to generate and refine candidate research questions. All questions were assessed and selected by the authors, who take full responsibility for the study design.”
Drafting. “Generative AI (Claude, Anthropic) was used to produce initial drafts of the introduction, which the authors then substantially rewrote, fact-checked, and referenced. The authors are responsible for all content.”
Editing. “A large language model (GPT-4o, OpenAI) was used to improve the clarity and grammar of author-written text. No content or citations were generated by the tool.”
Analysis support. “A local language model assisted first-cycle coding of interview transcripts; every suggested code was reviewed, accepted, or rejected by the authors, and the decisions were recorded. Interpretation and theme development were carried out by the authors.”
Each names a tool, a task, a stage, and — crucially — the human who stayed responsible. Swap in your own tool and task, and delete the ones that do not apply.
Checklist & templateAssemble yours
Downloadable · disclosure-sentence picker
The disclosure picker
- Name the tool and its version (e.g. GPT-4o, Claude, a local model).
- Name the task: brainstorming, drafting, editing, summarising, or analysis.
- Name the stage of the work where you used it.
- State the human review that followed — who checked the output.
- Check the target venue’s current AI policy for placement and threshold.
- Confirm no tool is listed as an author.
- Confirm the statement matches your own record of what happened.
The honest limit. A disclosure records what you did; it does not make the AI use appropriate — that depends on your venue’s policy and your own judgement. Its benefit is that a specific, matching statement leaves nothing for a reader to suspect.
Related guidesRead next
Recommended tool · the next step
AI Disclosure Kit
Assemble the disclosure sentence for your journal, from what you actually did — the AI Disclosure Kit keeps your AI-use records, disclosure text, and each venue’s requirements together, and gives you the wording to check and adapt before submission.
See the AI Disclosure Kit · €29 →Further readingSources worth your time
- ICMJE, Recommendations — use of AI-assisted technology — icmje.org.
- COPE, Authorship and AI tools — publicationethics.org.