# How to disclose your use of AI in a statistical analysis

> Name the tool and date, say what it did and what you checked, and put it where your journal asks. Current ICMJE, Nature Portfolio and Elsevier wording, fill-in templates and a log to keep.

Canonical: https://www.licklider.ai/guides/how-to-disclose-ai-use-in-statistical-analysis/
Section: Guides › Using AI for statistics
Author: Licklider · Published 2026-11-11 · Last reviewed 2026-11-11 (UTC)

> **Quick answer.** When an AI assistant helped with your statistics, write **which tool** you used (name, and the version or the date you used it), **what it did** (drafted code, suggested the test, explained output, edited text), **what you checked and how** (re-ran the code, checked the assumptions, confirmed the numbers in the paper match the output) and put that statement **where your journal asks**: ICMJE and Elsevier put analysis use in the Methods and writing help in the acknowledgments or a declaration section, and ICMJE also wants the use described in the submission cover letter. Write only checks you actually did. Keep authorship with the people who take responsibility for the work and describe the tool's role in the text instead: ICMJE, Nature Portfolio and Elsevier all rule out listing it as an author. Policies were checked on 10 November 2026; your journal's own page has the last word.

> **In ten minutes.** 1. Find your situation in the table under *Which section: Methods, acknowledgments or a declaration*. 2. Copy the matching template and replace every bracketed item. 3. Check that each claim in it ("re-ran", "checked the group sizes") is something you actually did; if you have not re-run the code yet, do that first ([how to check](/guides/how-to-check-an-ai-statistical-analysis/)). 4. Add the one-line cover-letter version: ICMJE journals require the use to be described in both the cover letter and the manuscript, and other journals may ask on their submission form. 5. Read your journal's AI policy page once more before you submit.

> **Terms used below.** *AI assistant*: a chatbot or language model (for example ChatGPT or Claude) that writes text or code from your prompts. *Disclosure statement*: the sentences in your paper that say how you used it. *Methods*: the section that describes how the data were collected and analysed. *Declaration section*: a separate statement some publishers ask for before the references. *Prompt*: what you typed to the assistant. *Re-run*: running the code yourself, in your own software, on your own data. *ICMJE*: the International Committee of Medical Journal Editors, whose recommendations many journals follow. *COPE*: the Committee on Publication Ethics. *Cover letter*: the letter sent to the editor with the first submission; the *response letter* is the point-by-point reply to reviewers sent with a revision. *Acknowledgments*: the section thanking people and services that are not authors. *Repository*: a public archive for code and data, such as Zenodo or OSF. *Welch's t-test*: compares two group means without assuming equal spreads. *Mixed model with mouse as a random effect*: a regression that accounts for several measurements per mouse. *Group sizes and units*: how many animals were in each group and whether each row is an animal.

## What the main policies ask

The policies agree on two points: disclose the use and keep authorship human. ICMJE and Elsevier also specify the Methods for analysis use; the current Nature Portfolio page names no section, and Nature's 2023 editorial said methods or acknowledgements. They differ in where writing help goes and in how they grade the risk (Nature Portfolio sorts uses into green, permitted; amber, permitted only with your verification and disclosure; and red, not permitted).

| Source (each re-checked 11 November 2026; your journal's own page has the last word) | What it says |
| --- | --- |
| [ICMJE Recommendations, Defining the role of authors and contributors](https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html), section 4 | "Authors who use such technology should describe, in both the cover letter and the submitted work in the appropriate section if applicable, how they used it. For example, if AI was used for writing assistance, describe this in the acknowledgment section (see Section II.A.3). If AI was used for data collection, analysis, or figure generation, authors should describe this use in the methods (see Section IV.A.3.d)." Also: "Authors should not list AI and AI-assisted technologies as an author or co-author, nor cite AI as an author." Its section IV.A.3 ([Preparing for submission](https://www.icmje.org/recommendations/browse/manuscript-preparation/preparing-for-submission.html)) adds: "Authors who used AI technology to conduct the study should describe its use in the methods section in sufficient detail to enable replication of the approach", "including the tool used, version, and prompts where applicable". |
| [Nature Portfolio editorial policy, Artificial Intelligence](https://www.nature.com/nature-portfolio/editorial-policies/ai) | A risk framework. "Recommending statistical tests or modelling approaches" and "Explaining outputs from statistical models in plain language" are in the amber level: "Permitted with human oversight, verification, and transparency through disclosure." Polishing language, "Comparing methodological options" and "Data cleaning and deduplication" are green: "Permitted. Disclosure enhances trust and transparency and clearly demonstrates human accountability." Red, not permitted, includes "Assigning authorship or accountability to AI systems or tools" and "Fabricating data, citations or results". Authors are expected to declare their AI use "for trust and transparency". |
| [Nature, editorial of 24 January 2023](https://www.nature.com/articles/d41586-023-00191-1) | Earlier wording, not the current policy (the current Nature Portfolio page names no section): "researchers using LLM tools should document this use in the methods or acknowledgements sections." For analysis use, follow ICMJE and Elsevier and use the Methods. |
| [Elsevier, Generative AI policies for journals](https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals) | "Where AI tools are used as part of the research process rather than manuscript preparation, this use should be described in detail in the Methods section." For writing help it recommends a section titled "Declaration of generative AI and AI-assisted technologies in the manuscript preparation process". Its FAQ adds that the policy "does not prevent the use of AI tools in formal research design or research methods, including but not limited to study design, code development and data analysis" — use that is still to be described in the Methods. On authorship: "Authors should not list AI tools as an author or co-author, nor cite AI tools as an author." |

The COPE position statement *Authorship and AI tools*, the Science journals' editorial policies and Cell Press author guidelines could not be opened for quotation on 10 November 2026, so they are not quoted here; if you submit to those journals, read their pages directly.

**Action:** open your journal's own AI policy now and note two things: which section it wants, and whether the cover letter must mention the use too (ICMJE says both).

## Which section: Methods, acknowledgments or a declaration

Decide by what the assistant did, not by which tool it was. This page's default is conservative: if the assistant touched the analysis in any way, the Methods carries the statement, because a reader needs it to judge the result — even where Nature Portfolio grades the use green. The table follows the policies as checked on 10 November 2026; ICMJE journals also want the use in the cover letter.

| What the assistant did | Where to describe it | Template | Why |
| --- | --- | --- | --- |
| Drafted or debugged analysis code | Methods | Situation 1 | Analysis use (ICMJE; Elsevier "code development and data analysis") |
| Suggested which test or model to use | Methods | Situation 2 | Amber at Nature Portfolio: permitted only with your verification and disclosure |
| Cleaned, transformed or reshaped the data | Methods | Situation 1, adapted: say the assistant "cleaned and reshaped the data" instead of drafting the code | It changed the data you analysed |
| Explained output so you could interpret it | Methods, briefly | Situation 4 | Amber at Nature Portfolio ("Explaining outputs from statistical models in plain language") |
| Edited language only, no analysis | Acknowledgments (ICMJE) or the declaration section (Elsevier) | Situation 3 | Writing assistance |
| Explained a concept while you learned, with no effect on this analysis | Not required by these policies; mention it if your journal's form asks about any use | None needed | It did not produce any part of the submitted work |

**Action:** pick every row that applies. Several can apply to one paper; then the Methods statement covers the analysis rows and the acknowledgments or declaration covers the editing row.

## Templates for your manuscript

Replace every item in square brackets; where a bracket offers options separated by slashes, keep the one that is true and delete the rest. Keep only sentences that are true for your study: if you have not yet re-run the code, run the [checks](/guides/how-to-check-an-ai-statistical-analysis/) first, because a disclosure cannot claim a check you did not do. Name the tool, the version or model if the interface shows one, and the date; if no version is shown, the name and date are enough. If you do not know the name of your analysis, describe it in plain words ("a comparison of liver weight between two groups of mice") and see the [test chooser](/tools/statistical-test-chooser/).

**Situation 1 — the assistant drafted code that you re-ran and checked (Methods):**

> Analysis code for [the comparison of group means / the mixed model / the figure] was drafted with [name of tool, version or model if shown] on [date] from prompts describing the planned analysis [prompts are in Supplementary File X, if your journal asks for them]. We re-ran the code in [software and version you ran], checked [the group sizes and units against the raw data / the software defaults that change results] and confirmed that every reported value matches the output [delete any check in this sentence that you did not do]. The final code is available at [repository, such as Zenodo or OSF / the supplementary files / from the corresponding author on request].

Filled in for a mouse study (tool name left as a placeholder): "Analysis code for the comparison of liver weight between groups was drafted with [tool name] on 3 November 2026 from prompts describing the planned analysis. We re-ran the code in R 4.3.3, checked the group sizes and units against the raw data, and confirmed that every reported value matches the output. The final code is available in Supplementary File 2."

**Situation 2 — the assistant suggested the test (Methods):**

> [Name of tool, version if shown] was used on [date] to compare candidate analyses for [the outcome] and suggested [the test it suggested]. The authors [accepted / changed] this and used [Welch's t-test / a linear mixed model with mouse as a random effect / the analysis in plain words] because [the design: one value per animal, two independent groups] [(reference)]. The analysis was [specified in the protocol of [date] before the data were analysed / chosen after the data had been seen]. [Name of statistician / The authors; no statistician was consulted] checked the choice against [the design / a statistics reference].

The timing sentence is not optional: if you cannot point to a dated protocol, registration or notebook entry that names the test, write "chosen after the data had been seen", and see [what counts as a planned analysis](/guides/reviewer-asks-for-multiple-comparison-correction/). None of these policies requires a statistician; say who checked.

**Situation 3 — language editing only (Acknowledgments, or the declaration section in Elsevier journals):**

> During the preparation of this work, the author(s) used [NAME OF TOOL / SERVICE] in order to [improve the readability of the Discussion]. After using this tool/service, the author(s) reviewed and edited the content as needed and take(s) full responsibility for the content of the published article.

This is Elsevier's recommended statement, quoted with the reason filled in as an example; Elsevier asks for it in a separate section immediately before the references. It covers writing help only — analysis use goes in the Methods (Situations 1, 2 and 4).

**Situation 4 — the assistant explained output you then interpreted (Methods, one sentence):**

> [Name of tool, version if shown] was used on [date] to explain [the output of the mixed model]; the interpretation reported here is the authors', checked against [the software documentation / a statistics reference].

**Cover letter (ICMJE journals, and any journal whose form asks):**

> An AI assistant ([name of tool]) was used to [draft analysis code / suggest the statistical test / edit the language]; its use is described in the [Methods for analysis use / Acknowledgments for language editing], and the authors re-ran and checked the analysis.

Delete "re-ran and checked the analysis" if the use was language editing only.

**Action:** paste the template, fill it, then read it against your code and notes: every verb ("re-ran", "checked", "confirmed") must describe something you did.

## What to record while you work

The statement is only as good as your notes. Keep a short log in the project folder, one row per session:

| Date | Tool and version shown | What you asked (short) | What it produced | What you did with it | What you checked |
| --- | --- | --- | --- | --- | --- |
| [2026-11-03] | [tool, model name] | [Code for two-group comparison of liver weight] | [Python using `ttest_ind`] | [Changed to `equal_var=False`; kept the rest] | [Re-ran; compared with R `t.test`; values match] |

Save the final code you ran, not only the assistant's draft, and record the software versions you used. Elsevier's FAQ suggests the same: "Authors should keep a separate record of which tool and model was used, and how AI tools were used." The log lets you answer a reviewer who asks exactly how the assistant was used, and it makes the Methods sentence easy to write.

No log from earlier sessions? Recover the code you actually ran, its saved output and the versions it printed, write the statement only for checks you can point to, and delete any clause you cannot confirm.

**Action:** start the log today; fill in the row for the session you remember least well first.

## Before you paste data into an assistant

Nature Portfolio's policy says: "Manuscripts, peer review reports, and sensitive data must not be shared with unsecured or public AI systems." Patient data, unpublished results under embargo, and data shared under a collaboration agreement fall under your institution's rules too. Use an institution-approved service or a de-identified extract, or use made-up data with the same structure only to draft the code — never as the data you analyse — and then run the code on the real data yourself.

**Action:** check your institution's guidance on approved AI services before your next session.

## Wording that reviewers and editors push back on

| Instead of | Write |
| --- | --- |
| "AI was used in this study." | Name the tool, the version or date, and what it did. |
| Listing the tool as an author, or citing it as one | Describe the use in Methods or acknowledgments; authorship stays with people who take responsibility (ICMJE, Nature Portfolio, Elsevier). |
| "The analysis was verified." | Say how: re-ran in [software], checked [assumptions], compared with [reference]. |
| A Methods statement in passive voice that hides who chose the test | "The authors chose [test] because [design reason]." |

**Action:** search your draft for "AI" and "ChatGPT" and check each mention against this table.

## If you later find the assistant's choice was wrong

Re-run the analysis with the right method, report the corrected result, and say what changed. Before submission, that is an ordinary revision of your Methods; after submission, tell the editor in the response letter that accompanies the revision.

Checking the analysis before you write the statement, as in [how to check a statistical analysis an AI assistant ran](/guides/how-to-check-an-ai-statistical-analysis/), catches most such problems early.

**If a reviewer asks how the assistant was used, or the use was not disclosed at submission,** paste a paragraph like this into the response letter (with your filled Situation 1, 2 or 4 text, without the quotation formatting) and add the Methods statement to the manuscript:

> We have added a statement to the Statistical analysis section (page [x], lines [y–z]): [your Situation 1, 2 or 4 statement]. [If the analysis changed: Re-running the analysis with [method] changed [result] from [old] to [new]; the revised values are in [table or figure].]

**Action:** if you are unsure whether a choice was right, follow the steps on the check page before you finalise the statement.

## References

- International Committee of Medical Journal Editors. Recommendations for the conduct, reporting, editing, and publication of scholarly work in medical journals: Defining the role of authors and contributors, section 4, Artificial Intelligence (AI)-Assisted Technology. [icmje.org](https://www.icmje.org/recommendations/browse/roles-and-responsibilities/defining-the-role-of-authors-and-contributors.html) (accessed 11 November 2026).
- International Committee of Medical Journal Editors. Recommendations: Preparing a manuscript for submission to a medical journal, section IV.A.3 (Methods). [icmje.org](https://www.icmje.org/recommendations/browse/manuscript-preparation/preparing-for-submission.html) (accessed 11 November 2026).
- Springer Nature. Nature Portfolio editorial policies: Artificial Intelligence (AI). A risk-assessment framework for responsible AI use in research publishing. [nature.com](https://www.nature.com/nature-portfolio/editorial-policies/ai) (accessed 11 November 2026).
- Tools such as ChatGPT threaten transparent science; here are our ground rules for their use. *Nature* 613, 612 (2023). [doi:10.1038/d41586-023-00191-1](https://doi.org/10.1038/d41586-023-00191-1) (earlier wording; the current Nature Portfolio policy names no section)
- Elsevier. Generative AI policies for journals. [elsevier.com](https://www.elsevier.com/about/policies-and-standards/generative-ai-policies-for-journals) (accessed 11 November 2026).

## Related

- [How to check a statistical analysis an AI assistant ran](https://www.licklider.ai/guides/how-to-check-an-ai-statistical-analysis/)
- [How to write the statistical methods section](https://www.licklider.ai/guides/how-to-write-statistical-methods-section/)
- [ChatGPT for statistical analysis: what it does well and what to check](https://www.licklider.ai/alternatives/chatgpt-for-statistical-analysis/)

## How this page was made

- Written and published by Licklider. Every policy statement is quoted from the publisher's or ICMJE's own page, opened on 10 November 2026 and re-checked on 11 November 2026; pages we could not open are named and not quoted.
- The templates are for your manuscript; replace every bracketed item with your own details.
- Separate review passes checked the policy quotations and readability before publication.