At the end of a manuscript sits one line: “The authors used AI tools in conducting this research.” Readers cannot tell whether the AI fixed a few commas or wrote the discussion. The editor emails to ask, and the paper loses two weeks. A vague statement protects nobody and invites more suspicion, not less.
A good disclosure is short. It just needs to answer the five questions a reader will ask.
The five elements
- Which tool: type and name, with version or date of use if the policy asks. Version details give context, because tool capabilities change quickly.
- For what: language editing, structural feedback, code drafting, translation, searching, summarising.
- Where: the whole paper, only the background section, only the analysis code.
- How it was checked: how the authors reviewed, verified sources, reran code or edited the output.
- Responsibility: a sentence confirming the authors take full responsibility for the content.
Missing elements 2 and 4 are the most common failings. Element 4 matters most: it shows the AI was a tool you controlled, not a substitute writer.
Where to put it
- Methods: when AI was part of the research process, such as data analysis, qualitative coding or screening literature.
- Acknowledgements or a dedicated declaration: when AI helped with writing or language.
- Cover letter: a brief mention if the journal requests it.
- Coursework and theses: follow the institution's or course's template; many require a declaration page at the front.
Always follow the location given in the author guidelines. If none is specified, use the methods when AI affected results and the acknowledgements when it only affected wording.
Model wording for common cases
| Case | Model statement |
|---|---|
| Language editing | “During manuscript preparation the authors used [tool] to check grammar and clarity in text they had written. The authors reviewed every suggestion and take full responsibility for the content.” |
| Translation | “The manuscript was written in Vietnamese and translated into English with the assistance of [tool]. The translation was reviewed and edited for technical terminology by the authors and a colleague fluent in English.” |
| Code drafting | “Initial R code for data cleaning and plotting was drafted with the assistance of [tool]. All code was reviewed, rerun and verified on simulated data by the authors and is available at …” |
| Literature search | “[Tool] was used to suggest search terms at an early stage. The formal search was conducted in [databases] using the strategy described in …; every cited source was read directly by the authors.” |
| Structural feedback | “The authors used [tool] to obtain feedback on the structure of the Discussion. All text was written by the authors; feedback was weighed as one would weigh comments from a draft reader.” |
These templates describe specific activities. Do not use the language-editing wording if the AI actually rewrote whole paragraphs.
Phrases to avoid
- “AI was used.” Says nothing.
- “AI was used minimally.” Unverifiable, and readers will wonder what minimal means.
- “No AI was used” when you relied on your word processor's rewrite feature. That feature is generative AI too.
- Listing an AI tool as a contributor in the author list or CRediT roles.
A before-and-after example
Before: “The authors used AI to help write the paper.”
After: “Two authors (J.P. and A.K.) used a generative AI assistant, in the version available in March–April 2026, to check the grammar of the Introduction and Discussion, which they had written in English, and to draft initial code for Figure 2. Every language suggestion was reviewed sentence by sentence; the code was rerun and checked against the source data. The tool was not used to generate scientific content, analyse data or search the literature. The authors take full responsibility for the content.”
The second version is three sentences longer but leaves the editor with no questions. The negative sentence (“was not used to…”) is especially useful: it marks out clearly which work was entirely the authors' own.
Coursework and theses
Many courses ask for a more detailed declaration than journals do, sometimes including the prompts used and excerpts of AI output. The purpose is to help markers assess your own contribution fairly. A useful structure:
- Tools used and when.
- Each specific task given to the AI, one per line.
- One representative prompt and how you used the response.
- Which parts of the work involved no AI at all.
- What you learned or changed after seeing the AI's feedback, if the task asks for reflection.
An AI use log makes disclosure painless
The easiest way to disclose accurately is to record as you go rather than reconstructing at submission. A simple five-column table is enough:
- Date.
- Tool.
- Task given to the AI, in one line.
- Part of the work affected.
- What you used and how you checked it.
After a few weeks the disclosure practically writes itself from the log. The log is also valuable evidence if your work is ever questioned.
Co-authors and a joint statement
In multi-author papers, each person may have used AI differently. Before submission, the corresponding author should ask every co-author what they used AI for and combine the answers into one statement. All co-authors should read and approve it, since the whole team shares responsibility.
Your next step: create the five-column log in your current project folder, fill in the AI uses you can still remember, and keep adding to it from today. When submission comes, the disclosure will be a matter of copying.
Câu hỏi thường gặp
What should an AI disclosure statement include?
The tool used, the tasks it performed, which parts of the work were affected, how the authors checked the output, and a statement that the authors take full responsibility for the content.
Where do I put an AI disclosure in a journal article?
Wherever the journal specifies. If it does not, use the methods section when AI affected the research process, and the acknowledgements or a separate declaration when it only helped with writing.
Do I need to disclose using AI just for grammar?
It depends: some journals exempt basic grammar checks, others require every tool to be named. If unsure, a short sentence is the safe choice.
Should I include the AI tool's version number?
Include it if the policy asks or if AI affected results, because tool capabilities change between versions. For language help, the tool name is usually enough.
Will disclosing AI use hurt my chances of publication?
Honest disclosure in line with the journal's policy is not in itself grounds for rejection. Failing to disclose and being found out is what causes serious problems.
