How to Disclose AI Usage in Your Thesis

Your thesis needs a clean AI record

A thesis is a formal research document. Your committee reads it as a record of what you did, how you did it, and what work counts as yours.

If you used ChatGPT to test wording, GitHub Copilot to draft code, DeepL Write to clean up sentences, or another AI tool at any stage, say so in plain language. That protects you. It also helps your supervisor and examiners judge the work fairly.

By April 23, 2026, many universities had moved from vague advice to written thesis and dissertation rules. The details differ, but the pattern stays stable: disclose the tool, describe the task, state the limits, and show that you kept control over the research and final text. The University of Georgia says generative AI use in theses and dissertations is prohibited unless the advisory committee authorizes it within an approved scope, and any approved use must appear in a statement in the thesis or dissertation. Illinois State says all generative AI use in the thesis or dissertation process must be disclosed to the student’s committee. KU Leuven requires PhD students to add a GenAI statement to the dissertation, even if they used AI only for language assistance. (grad.uga.edu)

That last point matters. A disclosure is not a confession. It is documentation.

If you want the broader background first, read Do I Need to Disclose AI Usage in My Paper? and How to Disclose ChatGPT Usage in Academic Papers. If you already know that your thesis needs a statement, this guide will help you write one that your committee can actually use.

Check your local rule before you draft anything

Start with your own university, then your department, then your supervisor.

Do not assume that "AI is allowed if disclosed." Some schools permit limited use with committee approval. Some ask for disclosure in the manuscript. Some require discipline-specific program rules. Some departments ban generative writing help for theses and dissertations.

The University of Georgia requires committee authorization before a student can use generative AI in a thesis or dissertation, and it requires disclosure inside the manuscript if the committee approves that use. Illinois State requires disclosure to the committee and says departments may impose stricter standards. The University at Buffalo does not impose one campus-wide thesis rule, but it now requires graduate programs to publish their own AI policies for dissertations, theses, and capstones by Fall 2026. (grad.uga.edu)

Some departments go much further. The University of Illinois Chicago History Department states that students may not use generative AI programs to complete a PhD dissertation or MA paper, while allowing narrow assistive uses like proofreading and some data-related tasks with acknowledgment. (hist.uic.edu)

That means one thing for you: check the current rule before you write your disclosure. Save the email or policy PDF. Use the exact date when you checked it.

If you will later turn a thesis chapter into a paper, also read AI Transparency Requirements for Journal Submissions and [[[AI Disclosure](/ai-disclosure-in-systematic-reviews-and-meta-analyses/)](/how-to-disclose-ai-use-for-neurips-icml-and-acl-submissions/) Policies by Major Journals](/ai-disclosure-policies-by-journal/). A thesis can pass under one rule and still need a different disclosure later.

Put the disclosure where your readers will look for it

Most thesis disclosures belong in one of three places.

The first option is front matter. This works well when the AI use was limited, easy to describe, and spread across the thesis. KU Leuven says PhD students must add a GenAI statement near the beginning of the dissertation. UGA also expects any approved generative AI use to appear in a statement within the thesis or dissertation. (arts.kuleuven.be)

The second option is an appendix. Use this when you need room for dates, prompts, chapter-by-chapter notes, or verification steps. An appendix also works well if your graduate school wants a short statement up front but your committee wants more detail on file.

The third option is the methods section. Put AI there when it shaped the research process itself. If you used an LLM for annotation, preprocessing, coding support, synthetic text generation, image analysis, or scoring, then the AI use is part of the method. Treat it like any other method choice. Describe it there.

In practice, many theses need two layers: a short front-matter statement and a fuller appendix. That keeps the main document clean while giving examiners enough detail.

If you want a structured appendix, generate an [AI Usage Card](/). It gives you a format that you can attach to the thesis or copy into it. For more examples, see [AI Usage Cards Examples and Templates](/ai-usage-cards-examples/).

Answer five questions and your disclosure will hold up

Most bad disclosures fail because they hide the scope of use.

A useful thesis disclosure answers five questions.

Which tool did you use?

Name the product and the provider. Add the model, version, or access route if you know it.

"ChatGPT" is thin. "ChatGPT, OpenAI, accessed February to March 2026" is better. If you used Claude through a university platform or Copilot inside VS Code, say that too.

What did you use it for?

Name the task, not the vibe.

"Used AI during writing" says nothing. "Used GitHub Copilot to draft parts of the Python preprocessing script for Chapter 4" gives your reader something concrete to judge.

Where did that happen?

Point to a chapter, appendix, workflow step, or date range.

This lets your committee separate grammar help from work that touches the thesis argument, literature review, methods, results, or code.

How did you verify the output?

This is the part students skip. It is also the part committees care about most.

Say whether you checked citations against original sources, reran analyses, tested generated code, compared summaries against the papers, or discarded false outputs. If the tool produced text that you revised heavily, say that too.

What work remained yours?

Make authorship visible.

State that you designed the study, selected the sources, interpreted the findings, and approved the final text, code, tables, and figures. Heidelberg tells doctoral researchers to disclose when, where, why, and to what extent they used AI tools, and it states that the doctoral researcher bears full responsibility for the dissertation’s content, structure, and scientific integrity. (graduateacademy.uni-heidelberg.de)

That last line often decides whether a disclosure sounds honest or evasive.

Committees worry about two things more than students expect

The first is sensitive data.

Do not upload participant data, unpublished results, full thesis drafts, or confidential material into public AI systems unless your institution explicitly allows it and the tool is approved for that use. UAB tells graduate students not to upload thesis or dissertation manuscripts into AI detection tools and warns about ownership and intellectual property risks. Heidelberg tells doctoral researchers to anonymize prompts unless they use a university-internal system. (uab.edu)

The second is the difference between language help and content creation.

Some institutions treat grammar assistance more leniently than drafting, summarizing, coding, or idea generation. KU Leuven even gives separate wording for students who used AI only for "mere language assistance." (arts.kuleuven.be)

Do not hide broader use under a proofreading label. If you used AI to shape chapter outlines, generate code, summarize papers, or suggest arguments, write that directly.

If your work sits in a field where AI use can affect the research itself, read a field-specific guide too. You may find AI Disclosure for Qualitative Research, AI Disclosure for Social Science Research, and AI Disclosure for NLP Research Papers more useful than generic advice.

Example disclosure statements

Use these as patterns. Do not copy them line for line.

Minimal use for brainstorming and language help

I used ChatGPT (OpenAI, accessed February 2026) during the planning stage of this thesis to brainstorm possible research questions and refine search terms for the literature review. I also used DeepL Write for language editing in the final draft. I verified all factual claims, citations, and quotations against the original sources. The research design, source selection, analysis, and argument are my own work.

Moderate use for coding and drafting support

I used GitHub Copilot while preparing Chapter 4 to draft parts of the Python code for data cleaning and preprocessing. I reviewed, tested, and revised all generated code before using it in the analysis. I also used ChatGPT (OpenAI, accessed January to March 2026) to suggest outline options for Chapters 2 and 3 and to improve sentence clarity in selected passages. I did not rely on AI-generated references or unverified factual claims. I take responsibility for the text, code, analysis, and interpretation in this thesis.

Extensive use where AI was both a research tool and the topic

This thesis evaluates large language models for automated essay scoring. The models discussed in Chapters 5 and 6 form part of the research object. Separately, I used ChatGPT to draft helper scripts for data formatting and to propose an initial structure for the related work chapter. I substantially revised all AI-assisted text and verified each cited source against the original publication. I designed the study, selected the evaluation criteria, ran the analyses, interpreted the results, and wrote the final argument. A full AI Usage Card appears in Appendix C.

That third example does one useful thing. It separates "AI as object of study" from "AI used in the workflow." Examiners need that line.

A thesis-friendly LaTeX template

If you write in LaTeX, keep the main statement short and move the details to an appendix.

Front-matter statement

\chapter*{Declaration of AI tool usage}
\addcontentsline{toc}{chapter}{Declaration of AI tool usage}
 
During the preparation of this thesis, I used the following AI tools:
 
\begin{itemize}
  \item ChatGPT (OpenAI, accessed February--March 2026) to brainstorm
  research questions and improve wording in selected passages.
  \item GitHub Copilot to draft parts of the Python preprocessing code
  used in Chapter 4.
  \item DeepL Write for grammar and phrasing suggestions in the final manuscript.
\end{itemize}
 
I reviewed and verified all AI-assisted outputs.
I checked citations against the original sources and tested all generated code.
I designed the study, interpreted the findings, and take responsibility for the
final text of this thesis.

Appendix with chapter-by-chapter detail

\appendix
\chapter{AI usage details}
 
\section*{Chapter 2: Literature review}
Tool: ChatGPT (OpenAI, accessed February 2026)
 
Task: Suggested alternative search terms and possible section headings.
 
Verification: I located, read, and cited the sources myself. I did not cite
AI output as a source.
 
\section*{Chapter 4: Data processing}
Tool: GitHub Copilot
 
Task: Drafted parts of Python code for preprocessing and file handling.
 
Verification: I reviewed all code, ran tests, and corrected errors before use.
 
\section*{Language editing}
Tool: DeepL Write
 
Task: Grammar and style suggestions in the final draft.
 
Verification: I accepted or rejected suggestions manually.

A simple log table you can maintain while writing

\begin{tabular}{p{3cm} p{3cm} p{7cm}}
Tool & Chapter & Use and verification \\
ChatGPT & Chapter 2 & Suggested search terms; I verified all sources manually. \\
Copilot & Chapter 4 & Drafted preprocessing code; I tested and revised all code. \\
DeepL Write & Full thesis & Grammar edits only; I accepted changes manually. \\
\end{tabular}

This small table solves a real problem. Six months later, when someone asks what you did in Chapter 3, you will not need to guess.

For more LaTeX help, see LaTeX Tutorial for AI Usage Cards and How to Use AI Usage Cards in Overleaf.

Add an AI Usage Card as an appendix

A lot of students leave disclosure until the week before submission. That is when the wording gets vague and the memory gets bad.

An AI Usage Card gives you a cleaner workflow. You can generate the card while you write, export it, and attach it as an appendix. If your school prefers plain text, copy the same content into your preface, acknowledgments, or methods section.

You can even include the PDF directly in LaTeX:

\usepackage{pdfpages}
 
\appendix
\chapter{AI Usage Card}
\includepdf[pages=-]{ai-usage-card.pdf}

That gives your committee a standard record. It also gives you something you can reuse later if you submit a chapter as a paper, revise the work for a journal, or describe your process in a grant proposal. If you want to see what a finished card can look like, read AI Usage Cards Examples and Templates.

What not to write

Bad disclosures usually fail in one of three ways.

The first failure is vagueness.

"AI tools were used in the preparation of this thesis."

That sentence hides the tool, the task, and the scope.

The second failure is understatement.

"I only used AI for minor edits."

If you also used AI to shape chapter outlines, summarize papers, generate code, or revise paragraphs, that line will look evasive.

The third failure is missing verification.

"ChatGPT drafted the literature review in Chapter 2."

Any examiner will ask the same question next: how did you check the claims, quotations, and references? If you do not answer that, your reader will fill in the blank.

A good disclosure sounds calm. It is specific. It does not hide behind soft wording.

The safest workflow is boring, and that is why it works

Check the current policy. Ask your supervisor if the rule leaves room for interpretation. Keep a running log. Write the disclosure before the final week. Then generate your AI Usage Card and attach it to the thesis or paste the text into the required section.

That is all you need.

If you want a reusable record that your supervisor, committee, and later journal editors can all read, generate your AI Usage Card now. It will save you time, and it will give your thesis the one thing every examiner wants: a clear account of what you did.