Learn · Research integrity
Write an AI-use disclosure that says what happened
Submission portals can sense fatigue and reveal one final required box. “AI was used” will not fill it well. A useful disclosure names the tool, version, task, timing, and human review, then follows the target venue’s current policy.
After this guide you will be able to:
- Judge when you have to disclose AI use, and when you don't.
- Write a specific statement instead of a vague one.
- Find and record what your target venue actually requires.
AI was used.Specifics, please.
When you have to disclose
It depends on the venue and on what you used AI for. Most policies exempt basic spelling and grammar tools, but require disclosure once AI touches drafting, images, data work, or a literature search. The catch is that venues disagree on the details, so "do I need to disclose?" has no single answer. The safe habit is to assume you disclose, then check your venue's rule and record it.
What a good disclosure contains
A vague statement is the problem, not the use. Name four things, plainly:
- The tool. Which AI system, by name.
- The version. The model or version you used, where you can.
- What you used it for. The specific task: language editing, code, a first-draft outline, image generation.
- When. The dates or the study phase, if the policy asks.
Then state that you, the author, checked and take responsibility for the output. That last clause turns a vague sentence into a useful disclosure.
Vague: "AI was used in the preparation of this manuscript."
Specific: "We used [tool, version] for language editing of the Discussion in March 2026. The authors reviewed all edits and are responsible for the final text. No AI was used to generate data, results, or references."
The faults that get noticed
Three recur. A statement too vague to tell the reader anything. A co-author's AI session nobody else knew to disclose. And an AI-suggested reference that no one checked, which is the one that actually damages a paper. The fix for all three is a dated record kept while you work, not a statement reconstructed from memory at submission.
The policies genuinely conflict
This is why it is hard, and it is not your fault. The rules disagree on specifics: some publishers want the tool, version, manufacturer, and dates; some require you to acknowledge any use at all; some exempt copy-editing entirely. General guidance from ICMJE and COPE sets the direction, but your journal's own instructions win. So check your venue's page, on the date you submit, and write down what it said.
Specific beats vague. A dated statement about a defined task reads as competence. What damages a paper is the hidden session, the unchecked citation, or language so broad that nobody can tell what happened.
Stop improvising the statement at submission.
You used AI somewhere, and now you have to disclose it under rules that keep changing across venues. The policies are fragmented, and most people write the statement from memory the night before they submit. The AI Disclosure Kit replaces that with a dated AI-use log you write the statement from, seven fill-in manuscript statements, thesis and grant templates, and a directory of 12 major publisher and funder policies checked against their own pages. It helps you document and decide. The journal's, funder's, and institution's rules always win, and what you disclose stays yours.
See the AI Disclosure Kit · €29Related guides
See also how to write a scientific paper and make your analysis code citable. More guides are on the Learn page.