AI tools for research: a disclosure log template for papers and theses

Do not wait until submission week

The worst time to reconstruct AI use is the night before you submit.

By then, you may remember that you used a chatbot to revise a paragraph. You may remember that you used a literature tool to screen abstracts. You may even remember the tool name. But you may not remember the version, date, prompt, settings, output, or whether the tool touched unpublished data.

That gap matters.

Many journal and publisher policies now ask authors to disclose AI use in the manuscript, cover letter, declaration section, methods, acknowledgments, or figure captions. ICMJE says journals should require authors to disclose AI-assisted technology use at submission, and authors should describe that use in the cover letter and in the submitted work when relevant. ICMJE also says chatbots and other AI tools should not appear as authors because they cannot take responsibility for the work. (icmje.org)

A simple AI disclosure log solves a boring problem before it becomes stressful.

This guide gives you a template you can use in a lab notebook, spreadsheet, Overleaf project, thesis folder, or shared project drive. It also shows how to turn that log into an [[[AI Usage Card](/ai-disclosure-for-social-science-research/)](/chatgpt-disclosure-academic-papers/)](/what-are-ai-usage-cards/) before you submit.

What counts as an AI tool for research?

For disclosure planning, treat "AI tool" broadly.

Do not limit the log to chatbots. Record any tool that generates, rewrites, classifies, summarizes, codes, translates, searches, extracts, ranks, labels, predicts, or creates content that may shape the research record.

That includes tools for:

Literature search and paper discovery.

Abstract screening.

PDF summarization.

Interview transcript coding.

Translation.

Statistical code generation.

Image generation or figure editing.

Manuscript drafting.

Language editing.

Reference checking.

Peer review support.

Data cleaning.

You do not need to panic over every spell checker. You do need a record when a tool changes research content, interpretation, wording, data handling, analysis, or review.

If you want a broader decision rule, start with Do I need to disclose AI usage in my paper?. If you want tool choice guidance before the project begins, pair this log with How to choose and disclose AI tools for research.

The disclosure log, in plain form

Use one row per AI-assisted task.

A row should let a future reader answer four questions:

What did you use?

What did you ask it to do?

What material did it process?

How did a human check the result?

That last question matters most. A tool log without human review reads like a receipt. A good log shows responsibility.

Use these fields:

This looks like admin work. It is. But it takes less than a minute when you do it during the project.

Copy this AI tool log template

You can copy this table into a markdown file called ai-tool-log.md.

# AI tool log
 
Project:
Authors:
Last updated:
 
| Date | Stage | Tool and provider | Version or model | Input material | Task | Output used | Human review | Changes made | Disclosure location | Privacy note | Record link |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 2026-08-03 | Writing | Tool name, provider | Model or version | Draft introduction, no data | Asked for sentence-level language edits | Accepted edited phrasing in two sentences | First author checked against original meaning | Rejected one suggested claim | Acknowledgment or AI disclosure statement | No personal or unpublished data uploaded | /records/2026-08-03-language-edit.md |

If your lab uses spreadsheets, keep the same columns. The format matters less than the habit.

For teams, store the log where all authors can see it. That prevents awkward submission-week emails where one co-author says, "I also used a tool for coding interview excerpts. Is that a problem?"

Maybe it is. Maybe it is not. But the group should know before submission.

Add a prompt record when the task affects content

You do not need to archive every spelling suggestion.

Save prompts and outputs when the tool shaped research content. That includes data labels, generated code, summaries of papers, interview coding suggestions, figure edits, or manuscript text that you copied into the draft.

Use a short record like this:

# AI prompt record
 
Date:
Project:
Tool:
Version or model:
User:
Project stage:
 
## Input material
 
Describe what you pasted, uploaded, or linked.
 
## Prompt
 
Paste the prompt here.
 
## Output used
 
Describe the part used in the project.
 
## Human check
 
Describe how the author checked the output.
 
## Decision
 
Accepted:
Modified:
Rejected:
Needs follow-up:

This record helps if an editor asks for more detail. It also helps you write a short and honest disclosure later.

For examples of finished statements, see AI Usage Cards examples and templates.

Match the log to the part of the paper

Different AI uses belong in different places.

Language editing often fits in an acknowledgment, declaration, or publisher-specific AI statement. A tool that helped design a search strategy may belong in methods. A tool that classified abstracts in a review needs clearer methods detail. A tool that created or modified an image may need a figure caption too.

Elsevier’s journal policy says authors who use generative AI in manuscript preparation must include a separate declaration section near the end of the manuscript, before the references. Its policy also gives special attention to AI use in images and artwork, including disclosure in the figure caption and manuscript AI disclosure statement for allowed cases. (elsevier.com)

Nature Portfolio journals allow generative AI tools for manuscript writing or readability editing when authors declare that use in the Methods section. Nature also warns peer reviewers not to upload manuscripts into generative AI tools because that may compromise confidentiality. (doi.org)

Those examples show why a log needs a "disclosure location" field. Journals do not all ask for the same placement.

If you submit to journals with changing policies, check AI disclosure policies by major journals and AI transparency requirements for journal submissions before you finalize the manuscript.

Example 1: literature search support

A literature search tool can shape the paper even if it never writes a sentence.

Suppose you use an AI tool to suggest search terms for a scoping review. You then edit those terms, test them in two databases, and document the final strings.

Your log entry might say:

The disclosure can stay short:

\section*{AI usage disclosure}
 
The authors used an AI-assisted search tool to suggest candidate search terms during the development of the literature search strategy. The authors reviewed, edited, and tested all search terms before use. The final database search strings appear in Appendix A.

For systematic reviews, keep more detail than you think you need. Screening, extraction, and synthesis affect reproducibility. See AI disclosure in systematic reviews and meta-analyses for a review-specific version of this workflow.

Example 2: qualitative coding support

Qualitative researchers face a different problem.

AI tools can suggest codes, summarize interviews, translate excerpts, or group themes. That can help with exploration, but it can also flatten context. It can miss irony, local meaning, or power dynamics in the data.

A log entry should name the data type and the human check.

A methods sentence could read:

During exploratory coding, the authors used an AI-assisted text analysis tool to generate provisional code suggestions for five anonymized interview excerpts. Two researchers checked all suggestions against the full transcripts, revised the code labels, and made the final coding decisions.

If your project uses interviews, fieldnotes, focus groups, or ethnographic material, read AI disclosure for qualitative research before you upload anything to an external tool.

Example 3: code generation for analysis

Many researchers use AI tools to draft R, Python, Stata, Julia, or MATLAB code.

The disclosure problem does not end with "we used AI for code." Editors and readers need to know whether the tool wrote analysis code, comments, tests, visualizations, or only helped debug syntax.

A useful log entry records the script name and the validation step.

A repository note could read:

\section*{AI assistance in code preparation}
 
The authors used an AI coding assistant to suggest edits to plotting scripts. The authors tested all code, inspected the outputs, and take responsibility for the analysis and figures.

That wording will not fit every paper. It gives you a starting point.

Example 4: manuscript language editing

Language editing causes the most confusion because it can feel minor.

If a tool only corrected grammar, spelling, or sentence flow, many journals still want disclosure. Others draw a line between standard editing tools and generative AI. Your log protects you either way.

Record the section, task, and whether the tool changed meaning.

A short disclosure could read:

\section*{Declaration of AI-assisted writing}
 
The authors used an AI writing assistant for grammar and clarity edits in selected parts of the manuscript. The authors reviewed and edited all output and take responsibility for the final text.

If you need a tool-specific disclosure, see the guides on Claude, Google Gemini, and Microsoft Copilot.

Use the log before the cover letter

Before submission, scan the log and sort entries into four groups.

First, entries that do not need disclosure because the tool had no effect on the research record.

Second, entries that belong in the acknowledgment or AI declaration.

Third, entries that belong in methods because they affected design, data handling, analysis, screening, coding, or synthesis.

Fourth, entries that raise policy or ethics questions. These may involve personal data, confidential peer review material, patient information, third party manuscripts, copyrighted corpora, or generated images.

This sorting step turns a messy project history into a clean disclosure.

It also helps the corresponding author. Submission systems often ask one person to confirm AI use for the whole team. That person should not guess.

If you work with conference submissions, add the disclosure check to your deadline plan. AI conference deadlines and disclosure planning gives you a way to build that review into the last week before submission.

Turn the log into an AI Usage Card

A log stores project detail. An AI Usage Card turns that detail into a readable disclosure record.

Use the log to fill in the card fields:

Tool name.

Purpose.

Research stage.

Input type.

Output type.

Human oversight.

Disclosure text.

Limitations.

Privacy notes.

Then generate the card at ai-cards.org. You can export the result as text for a manuscript, keep it as a project record, or include it in supplementary material.

If you write in LaTeX, you can add a compact card section like this:

\section*{AI Usage Card}
 
\begin{description}
  \item[Tool] Name of AI tool and provider.
  \item[Purpose] Grammar editing and code debugging.
  \item[Research stage] Writing and analysis.
  \item[Input] Draft manuscript paragraphs and synthetic sample data.
  \item[Output] Suggested sentence edits and code revisions.
  \item[Human oversight] The authors reviewed, edited, tested, and verified all outputs.
  \item[Disclosure] The authors used AI assistance as described above and take responsibility for the final manuscript.
\end{description}

For more LaTeX options, use the LaTeX tutorial for AI Usage Cards or the Overleaf guide.

A small habit beats a perfect memory

You do not need a complex governance system to disclose AI use well.

You need a running record.

Write down the tool, the task, the input, the output, and the human check. Do it while the work happens. Your future self, your co-authors, and your editor will thank you.

Before you submit your next paper, thesis, review, or grant proposal, open your AI tool log and generate an AI Usage Card at ai-cards.org. Use the card as a disclosure statement, an appendix item, or a project record that shows what you used and how you checked it.