How to Choose and Disclose AI Tools for Research

The best AI tool is the one you can explain

Many researchers search for the best AI tools for research. Fair enough. The pressure is real: read faster, write clearer, code cleaner, find gaps sooner.

But a better question comes first.

Can you explain what the tool did, what you checked, and what part of the final work changed because of it?

That question matters more than any ranking. A tool that helps you fix commas may need no detailed note. A tool that helps you screen literature, generate code, classify interview excerpts, or rewrite claims in a results section needs a record you can defend.

If you want the short version: choose tools by task, risk, and traceability. Then generate an [[AI Usage Card](/chatgpt-disclosure-academic-papers/)](/) before submission so your disclosure does not depend on memory.

start with the task, not the tool

Do not begin with the brand name.

Begin with the work.

Ask what you need help with:

• finding literature
• summarizing papers
• translating text
• editing language
• writing code
• debugging analysis scripts
• classifying data
• drafting text
• creating figures
• preparing a peer review

Each task changes the disclosure problem.

Language editing affects presentation. Literature screening can affect the evidence base. Code generation can affect results. Classification of qualitative data can affect interpretation.

That is why a single sentence like "AI was used during preparation" rarely gives enough information. It hides the part that readers care about.

If you need a fuller starting point, read What Are AI Usage Cards?. The card format asks for the tool, purpose, input, output, and human checks. That sounds basic. In practice, those five details solve many disclosure headaches.

sort AI use by research risk

A simple risk sort helps you decide how much to record.

Low risk use usually changes wording, formatting, or spelling. You still own the meaning. Examples include grammar checks, reference formatting checks, or rephrasing a sentence after you wrote the content.

Medium risk use helps you think, search, or organize. Examples include summarizing a set of papers, suggesting search terms, comparing abstracts, or explaining unfamiliar methods. These uses may not appear in the final text, but they can shape your reading and framing.

High risk use touches evidence, analysis, or claims. Examples include screening studies for a review, coding interview data, generating analysis scripts, interpreting model outputs, creating images, or drafting result statements. Readers, reviewers, and editors have stronger reasons to know about these uses.

Journal guidance often follows the same logic, even when journals use different wording. ICMJE says journals should require authors to disclose AI assisted technologies used in producing submitted work, and it also states that authors should not list AI tools as authors. (icmje.org) CSE guidance tells authors to disclose use of AI tools and says human authors remain accountable for manuscript accuracy and attribution. (csescienceeditor.org)

If you want help deciding whether your use needs disclosure, see Do I Need to Disclose AI Usage in My Paper?.

use a five question check before you adopt a tool

Before you bring an AI tool into a research workflow, answer five questions in plain language.

First: what task will the tool perform?

Second: will the tool see confidential, personal, unpublished, copyrighted, or embargoed material?

Third: will its output affect data, analysis, interpretation, or claims?

Fourth: can a human author verify the output?

Fifth: can you describe the use later without exposing sensitive material?

If you struggle to answer one of these, pause.

This matters most for peer review and unpublished manuscripts. Nature Portfolio tells peer reviewers not to upload manuscripts into generative AI tools and asks reviewers to declare AI support if an AI tool helped evaluate manuscript claims. (nature.com) If you review papers, read [[[AI Disclosure](/how-to-disclose-microsoft-copilot-use-in-academic-writing/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) in Peer Review: What Reviewers and Editors Should Report](/ai-disclosure-in-peer-review-what-reviewers-and-editors-should-report/) before you paste any confidential text into a tool.

keep an AI use log while you work

Do not wait until submission week.

You will forget the model name. You will forget whether you used the tool for the introduction or the discussion. You will forget that one prompt that helped you rewrite a code comment into a method paragraph.

Keep a small log.

This log does not need to become an appendix in full. It gives you the raw material for a clear disclosure.

It also protects coauthors. Nobody wants to discover, after acceptance, that one author used a tool in a way the group cannot reconstruct.

know what policies usually ask for

Policies differ. You must read the instructions for your target journal, conference, or thesis office.

Still, several patterns repeat.

Many policies ask you to name the tool. Some ask for the version or model. Some ask where you used it. Some ask you to state that human authors reviewed and take responsibility for the final work.

JAMA Network guidance says authors should report the name of the AI tool, version, extension numbers when available, manufacturer, date of use, and a description of how they used it when AI created, reviewed, revised, or edited manuscript content. (jamanetwork.com) Elsevier allows generative AI and AI assisted technologies during manuscript preparation when authors apply oversight and disclose the use according to author instructions. (elsevier.com) Nature Portfolio asks authors to document LLM use in the Methods section or another suitable section, while it excludes AI assisted copy editing from that disclosure requirement. (nature.com)

For a wider policy view, use AI Disclosure Policies by Major Journals and AI Transparency Requirements for Journal Submissions before you submit.

place the disclosure where readers will look

You have four common locations.

Use the Methods section when AI affected research procedures, data processing, analysis, coding, study selection, classification, or figure generation.

Use the Acknowledgments section when AI helped with writing, editing, translation, brainstorming, or organization.

Use a dedicated AI disclosure statement when the journal offers one or when your AI use spans several parts of the work.

Use supplementary material when the workflow needs more detail than the paper can carry, such as prompts, logs, code review notes, or screening instructions.

For theses, check your graduate school rules. A thesis often gives you more space than a journal article, so you can add a short AI Usage Card or an appendix. See How to Disclose AI Usage in Your Thesis for thesis wording.

write the disclosure like a methods note

A good disclosure says what happened. It does not apologize. It does not brag.

Weak:

"AI tools were used to improve this manuscript."

Better:

"The authors used an AI writing assistant on 2026-04-12 to edit author written sentences in the abstract and introduction for grammar and clarity. The authors reviewed all suggested changes and retained responsibility for the final text."

For code:

"The authors used an AI coding assistant in April 2026 to suggest R code for sensitivity analyses. The authors modified the code, inspected it manually, and reran all analyses before including the results."

For literature work:

"The authors used an AI search assistant to suggest candidate search terms during protocol development. The final search strategy was written by the authors and reviewed against the inclusion criteria before database searches."

For qualitative work:

"The authors used an AI tool to propose preliminary labels for 20 pilot excerpts. The research team revised the codebook, coded the full dataset manually, and did not use AI generated labels as final codes."

If you work with interviews, field notes, or focus group transcripts, read AI Disclosure for Qualitative Research before you upload material to any external system.

add the statement in LaTeX

Many researchers need wording that fits a manuscript file, not a policy memo.

You can add a short disclosure section like this:

\section*{AI usage disclosure}
 
The authors used an AI writing assistant on 12 April 2026 to edit
author-written text in the abstract and introduction for grammar and
clarity. The tool did not generate research data, analysis, figures, or
scientific claims. The authors reviewed all suggested changes and take
responsibility for the final manuscript.

For Methods use, make the disclosure more procedural:

\subsection*{Use of AI-assisted tools}
 
During development of the analysis code, the authors used an AI coding
assistant to suggest draft R functions for a sensitivity analysis. The
authors edited the suggested code, checked it against simulated data,
and reran the analysis from the raw data files. The final code is
included in the supplementary repository.

If you use Overleaf, you can pair this with the workflow in How to Use AI Usage Cards in Overleaf. If you need a longer card in a supplement, use the examples in AI Usage Cards Examples and Templates or the LaTeX Tutorial for AI Usage Cards.

generate a card before the final submission check

An AI Usage Card works best when you create it before the submission portal opens.

At that point, you still remember the details. You can ask coauthors to confirm the wording. You can decide whether the paper needs a short disclosure, a Methods note, a supplement, or all three.

The card also helps when a journal asks a narrow question in the submission form. You may answer the form in one sentence, then keep the full card for your records or submit it as supplementary material if the journal allows it.

For example, a systematic review team might generate a card that records AI help with search term expansion, duplicate screening tests, and language editing. The manuscript disclosure can stay short, while the card preserves the workflow. If you conduct reviews, see AI Disclosure in Systematic Reviews and Meta-Analyses for details that matter in evidence synthesis.

choose tools you can defend

Research AI tools will keep changing. Names will change. Features will change. Pricing pages will change.

Your disclosure practice should not depend on any one tool.

Pick tools that let you protect confidential material, export or record useful details, verify outputs, and explain your workflow to a reviewer without embarrassment. If a tool helps but leaves no trace, creates claims you cannot check, or requires sensitive data you should not share, choose another route.

Then document the use while the work is still fresh.

Generate an AI Usage Card at ai-cards.org for your current paper, thesis, review, or grant draft. Use it as a submission note, a supplement, or the basis for the AI disclosure text in your manuscript. A few minutes now can save a confused exchange with an editor later.