How to Disclose AI Use in Academic Writing

AI can help you brainstorm, reorganize notes, or polish a sentence, but an honest academic submission needs more than a vague note that says “AI was used.” A useful disclosure tells the reader what the tool did, where it affected the work, and what you checked yourself.

There is no single statement that fits every school, course, journal, or assignment. Start with the rules that apply to your work, then document your process while you can still remember it. This guide gives you a practical way to do both.

Disclosure Is Not the Same as Permission or Citation

Three questions are often mixed together:

1. Was this use permitted? Your assignment, instructor, department, or institution decides that.
2. Should the use be disclosed? The applicable policy decides when and how you should describe the assistance.
3. Does a source need a citation? Citation gives credit for words, ideas, data, images, or other material you used.

A disclosure does not turn prohibited AI use into permitted use. It also does not replace citations to the sources behind your claims. The Naval Postgraduate School's guidance on citation and disclosure treats these as related but separate responsibilities.

If you need a broader review of allowed uses, verification, and authorship, begin with this responsible AI editing workflow for students. Then use the steps below to create a disclosure that matches what you actually did.

Step 1: Check the Exact Rule Before You Write

Read the assignment instructions and syllabus first. If they do not answer your question, check department or institution guidance. Policies can differ between courses and even between tasks in the same course.

Ask these questions:

• Is generative AI allowed for this assignment?
• Are some uses allowed, such as proofreading, while drafting is not?
• Must I obtain approval before using a tool?
• Is a disclosure, citation, appendix, prompt log, or all four required?
• Does the instructor provide required wording or a placement rule?
• Are there restrictions on entering unpublished, personal, or confidential material?

Princeton's student guidance recommends checking course and departmental rules and asking the instructor when expectations are unclear. That is safer than guessing from a general university page.

Record the rule you relied on. A link, screenshot, or dated note can prevent confusion if instructions change later.

Step 2: Keep a Small AI-Use Record

Do not wait until submission day to reconstruct your process. Keep a short record whenever AI materially affects the work.

You do not need to paste sensitive information into this record. Describe the task without repeating private data. If the policy requires full prompts or output, keep them in the approved location rather than a public document.

The University of Auckland's student guidance asks students to identify the tool, purpose, prompts, evaluation, and integration of the output. A simple running log makes those details much easier to supply accurately.

Step 3: Build a Five-Part Disclosure

A practical disclosure can be built from five pieces. Use only the detail relevant to your actual process and the applicable rule.

1. Identify the tool

Name the product and, when available, the model or version. Include the date if the tool changes frequently or the policy asks for it.

2. State the task

Describe the job you gave the tool: generating topic options, testing an outline, revising sentence clarity, explaining code, or checking grammar. “Helped with my paper” is too broad to be informative.

3. Locate the effect

Say which part of the work was affected. This could be the initial outline, the wording of two paragraphs, a data-cleaning script, or the entire draft's language review.

4. Explain your human checks

State what you did after receiving the output. For example, you may have compared claims with original sources, tested code, rejected suggestions, restored technical qualifiers, or rewritten passages in your own words.

For a repeatable accuracy pass, use this guide to fact-check AI-generated text before you use it. “I reviewed it” is less useful than naming the checks you performed.

5. Define the boundary of responsibility

Make clear what the tool did not do and what you remain responsible for. You might say that the analysis, source selection, interpretation, and final wording are your own. Only make that statement if it is true.

The CDC's current considerations for generative AI in scientific work similarly emphasize identifying the tool and purpose, the content affected, and the human oversight applied.

Step 4: Match the Detail to the Type of Use

Not every use needs the same explanation. Follow the required policy first, then make the statement proportionate to the tool's influence.

Language editing

If a tool only suggested grammar or clarity changes, identify that limited role and explain that you reviewed every change. If preserving precise meaning matters, compare the revised version with your source notes or original argument. This workflow for humanizing AI text without changing its meaning can help with that comparison.

Brainstorming and outlining

State that the tool generated options or organizational suggestions. Note which ideas you selected, changed, or rejected. Do not imply that brainstorming had no effect if it shaped the final argument.

Drafted prose, code, analysis, or media

When generated material appears more directly in the submission, give a fuller account. Identify the affected sections and the validation you performed. For code, say how you tested it. For factual prose, name the source-checking process. For images or data, follow any additional authorship and rights rules.

Step 5: Put the Statement in the Required Place

The correct location depends on the context. An instructor may request a note at the end of the assignment, a footnote, an appendix, an acknowledgment, or a separate cover sheet. A journal may require a methods section or submission form.

Do not hide a required disclosure in an unrelated bibliography entry. Do not invent a location if the policy specifies one. When there is no stated location, ask the instructor or editor; if that is not possible, use a clearly labeled note that readers can find.

Worked Examples: Vague Versus Useful

These examples are patterns to adapt, not universal required wording.

Too vague

I used AI to help with this assignment.

The statement does not identify the tool, task, affected material, or review.

Language-editing example

I used [tool and version] on [date] to suggest grammar and clarity edits in the final draft. I reviewed each suggestion, rejected changes that altered technical meaning, and verified the final claims and citations against the original sources. The argument, evidence selection, and final wording are my responsibility.

Brainstorming example

I used [tool and version] to generate possible questions and outline structures before drafting. I selected and substantially revised the outline, wrote the paper from the assigned sources, and did not use generated citations. The source analysis and final text are my own.

Code-assistance example

I used [tool and version] to suggest an approach for the data-cleaning function in Appendix A. I rewrote the returned code, tested it against the documented cases, and manually reviewed the transformed data. The analysis and interpretation were completed by me.

Specific language is credible because it can be checked against your record. Do not copy an example that describes work you did not perform.

What a Disclosure Cannot Fix

A clear statement improves transparency, but it cannot repair every problem. It cannot:

• make a prohibited use acceptable;
• verify a false claim or fabricated citation;
• give permission to upload confidential material;
• replace attribution for borrowed words or ideas;
• prove that you understood work you cannot explain;
• transfer responsibility for the final submission to the tool.

If the use violated a rule, stop and speak with the instructor or editor rather than trying to phrase the problem away.

Final AI Disclosure Checklist

Before submitting, confirm that you have:

• checked the rule for this specific assignment or publication;
• identified the tool, model or version when available, and date if relevant;
• described the task and the material it affected;
• explained the checks, revisions, or tests you performed;
• kept citations to human-authored sources separate from the AI disclosure;
• placed the statement where the policy requires;
• retained any prompt or output record the policy asks for;
• removed private or confidential information from inappropriate systems;
• used wording that matches your real process;
• accepted responsibility for the final work.

The best disclosure is not the longest one. It is the one that lets a reader understand the tool's actual role, the boundaries of that role, and the human judgment behind the submitted work.