Using AI tools in research writing responsibly

Generative AI tools are now part of many researchers’ daily work. People use them to polish English, suggest a structure for a chapter, summarise long documents, draft code, or brainstorm research questions. Used carefully, they can save time and help researchers who write in a second or third language. Used carelessly, they can introduce invented references, distorted findings, breaches of confidentiality and questions about who actually wrote the work.

Policies are still evolving, and they differ between journals, publishers, funders and universities. This guide sets out principles that hold across most of them, and practical steps to protect yourself and your readers.

The principle that does not change: you are responsible

Whatever tools you use, the named authors are accountable for every word, figure, citation and claim in the work. The Committee on Publication Ethics (COPE) has stated that AI tools cannot be listed as authors, because they cannot take responsibility for the submitted work, cannot declare conflicts of interest, and cannot manage copyright or licence agreements. The ICMJE (International Committee of Medical Journal Editors) recommendations take the same position: AI tools should not be listed as authors, authors should disclose at submission whether and how they used AI, both in the cover letter and in the relevant section of the paper, and AI-generated material should not be cited as a primary source. Many major publishers have adopted policies along the same lines. In practice, this means:

• Never list an AI tool as an author or co-author.
• Do not cite an AI tool as if it were a source of authority for a factual claim.
• Check everything an AI tool produces before it goes into your work.

Lower-risk uses

Most policies are relatively relaxed about uses that support your own writing without creating content or analysis. These typically include:

• Language editing: grammar, spelling, clarity and readability of text you have written.
• Restructuring suggestions: asking for feedback on the order of sections or the flow of an argument.
• Brainstorming: generating possible search terms, counter-arguments or questions you might have missed.
• Coding help: drafting or debugging analysis scripts, which you then test and understand fully.

Even here, some journals ask you to disclose the use, so check before you submit.

Higher-risk uses

Other uses carry real risks to accuracy and integrity, and many journals restrict or prohibit them:

• Generating substantive text such as literature reviews, discussion sections or conclusions. The output may be generic, inaccurate or close to existing published text.
• Finding or generating references. AI chatbots are known to produce citations that look real but do not exist, or that attribute claims to the wrong source. Never include a reference you have not located and read yourself.
• Summarising sources you have not read. Summaries can omit caveats, merge studies or invent findings.
• Analysing or interpreting data without verifiable, reproducible methods.
• Creating or altering images and figures. Many publishers do not allow AI-generated or AI-manipulated images in research papers, except where the use is itself part of the research and is described in the methods.

Confidentiality and data protection

When you paste text into an online AI tool, you may be sharing it with a third party, depending on the tool’s terms and settings. This matters in several situations:

• Peer review. Manuscripts under review are confidential. Many publishers explicitly tell reviewers and editors not to upload manuscripts, or parts of them, into generative AI tools.
• Research data. Interview transcripts, health records or any personal data should not be entered into tools that are not approved for that purpose by your institution and your ethics approval.
• Unpublished ideas. Think before pasting a grant proposal or unpublished findings into a public tool.

How to disclose AI use

Disclosure requirements vary, but a good disclosure is specific and brief. It usually states:

• Which tool you used (name and, if known, version).
• What you used it for, for example “to improve the grammar and readability of the manuscript”.
• That the authors reviewed and edited the output and take full responsibility for the content.

Journals differ on where this goes. Some ask for it in the methods section, some in the acknowledgements, and some in a separate declaration at submission. Follow the journal’s instructions. If a journal has no stated policy, disclosing anyway is the safer choice.

A practical workflow

1. Check the rules first. Read the policy of your target journal, your university and, if relevant, your funder.
2. Do the thinking yourself. Your research question, argument, analysis and interpretation should be your own.
3. Use AI on your drafts, not instead of them. Write first, then use a tool for language or structure feedback if permitted.
4. Verify every factual claim and reference against the original source.
5. Keep a record of what tools you used and for what, so you can disclose accurately and answer questions later.
6. Agree on AI use with co-authors. Everyone on the author list shares responsibility, so make sure all co-authors know what was used.
7. Disclose as required.

A note for students and supervisors

Universities are setting their own rules for theses, dissertations and coursework, and these can be stricter than journal policies. Students should ask their supervisor and check written guidance before using AI tools on assessed work. Supervisors can help by being explicit about what is allowed in their group, rather than leaving students to guess.

AI-detection tools exist, but their results are not reliable enough to settle questions of misconduct on their own. Clear records of your drafting process, notes and earlier versions are the best evidence of your own work.

How CollabCite fits in

Good collaborators are still the best check on a manuscript: a co-author who reads every line will catch what a tool gets wrong. CollabCite helps researchers find co-authors by field and research interest, and our Integrity Policy applies to everyone on the platform. Explore more in our resources or join as a founding member.