AI use disclosure statements, with examples

An AI use disclosure statement names the tool, explains its contribution, identifies the affected work, and describes your checks. Include the model or version when known, and follow your instructor’s, publisher’s, or client’s wording and placement rules. Adapt the examples below to your work. They are not officially required declarations.

Check the policy for this piece of work

Read the assignment brief, journal instructions, or agreed client requirements. Check what’s allowed, what you must report, and where the statement belongs. Keep those instructions with the project. A disclosure does not make a prohibited use acceptable.

Cornell’s teaching guidance offers instructors options for courses that allow AI: citations, explanations, reflections, and transcripts. For an explanation, it suggests recording the tool, date, model, purpose, affected work, prompts, and verification. Use these drafting questions alongside your course’s policy.

Elsevier’s journal policy allows manuscript assistance, including organization and literature-related tasks, under human oversight. When disclosure is required, submit a separate declaration describing the tool, its purpose, and your oversight. In those journals, describe AI used in the research itself in Methods. Follow the target journal’s instructions and declaration format.

If instructions are unclear, ask a specific question: “Should the statement cover outline suggestions as well as text revision?”

Gather the facts before choosing the wording

Record the following while the conversation and draft are open:

• The tool and model or version actually used, if known.
• The date or date range of use.
• The task: outlining, drafting, editing, coding, analysis, or another specific contribution.
• The parts of the work affected.
• The checks and revisions you completed, and where you keep the supporting record.

“Helped with writing” might mean suggesting three headings or drafting most of a report. “Suggested the order of the background section” is clearer. Name the sources, calculations, or passages you actually checked instead of simply saying “reviewed.”

Don’t guess missing details. A model available today doesn’t tell you which one you used last month. If you cannot recover a required detail, explain what is missing or ask how to record it.

Four original examples you can adapt

Replace the bracketed fields, remove claims that don’t describe your use, and add any missing assistance. Describe only the checks you’ve completed. Note outstanding checks in your working draft and finish them before claiming completion.

School assignment: planning an essay

“I used Claude, [model if known], on [date] to suggest an order for the comparison in [assignment section]. I used [specific suggestions] in my outline. I checked [specific points] against [course materials or sources I read]. The prompts and replies are in [appendix or transcript filename].”

Use this for an outline task. If you also kept sentences Claude supplied, describe that drafting separately from planning.

Research manuscript: revising the presentation

“While preparing [manuscript title], we used Claude, [model if known], during [date range] to propose changes to [named sections or features]. We accepted [description of changes] and checked them using [checks actually completed]. The conversation record is kept in [approved project location or supplement].”

Name the revision: paragraph order, sentence structure, terminology, or another task. If several authors accepted different suggestions, identify the affected sections before agreeing on the final wording.

Research workflow: assistance with code

“We used Claude, [model if known], during [date range] to help with [specific coding task] in [script or analysis stage]. The generated suggestions were [describe what was kept or changed]. We checked the resulting code using [checks or tests actually performed]. The prompts, responses, and relevant code versions are documented in [record location].”

Identify the code that produced the results and the checks you ran, with the detail your methods require. The research workflow guide covers keeping the conversation and research files together.

If you only ran the script on a small sample, describe that limit. Don’t turn a successful test run into a claim that every output, edge case, or scientific interpretation was independently verified.

Client work: preparing a report

“I used Claude, [model if known], during [date range] to help prepare [named part of the deliverable] by [specific task]. Its contribution to the delivered version was [description]. I reviewed [material] using [checks actually completed]. The supporting conversation is [attached or kept in the agreed location].”

Match the agreed scope. For a monthly performance report, specify whether Claude helped with the summary, headings, or analysis. Name any remaining limitation the recipient needs to know.

If your spreadsheet supplied the figures and Claude drafted the commentary, say so. Include any calculations it suggested too.

Back the statement with the right conversation record

Keep original prompts and replies together, with your explanatory note separate. If several chats contributed, make an index connecting each file to the relevant part of the work.

A client report might have one chat for the opening summary and another for a chart explanation. Names such as report-summary-transcript.pdf and report-chart-transcript.pdf distinguish them better than two files called Claude.pdf. Record known conversation dates separately from export dates.

In desktop Chrome, DocMyChat saves the open claude.ai conversation as PDF, Word, or Markdown. The export follows the displayed branch and includes the model if known. Save each relevant chat separately and check that it contains the exchange your statement describes.

Use the full conversation when the required record includes prompts and revisions. A single reply shows the output, without decisions made in the surrounding exchange.

You can keep a transcript without making it public. Use the required submission or storage route. For a link, check that the recipient can open the correct record. For a file or appendix, follow how to attach the conversation or the PDF export steps.

Keep disclosure and citation connected

MLA’s guidance on describing AI use calls for explaining significant AI assistance as well as citing sources. A reference identifies a response. A disclosure explains how it influenced your work. Your instructions may require both.

Use the Claude citation guide for APA, MLA, or Chicago formatting. Cite underlying research sources where they support your claims. Naming Claude does not identify which study, document, or dataset supports a statement.

Check every claim in the finished statement

Compare the statement with the final deliverable and saved conversations. Does it cover the relevant tasks and affected sections? Can you explain every claimed check?

Remove placeholders, check the required placement, and update the disclosure after any further AI assistance so it describes the submitted version accurately.

Questions