Grad Students: Disclosing AI Assistance in Your Thesis or Dissertation

A thesis isn’t a problem set. It gets bound, archived, examined in person, and sometimes mined for journal articles years later, so the question of whether to mention the AI tools you leaned on carries more weight than it would for a weekly essay. The good news: a clear disclosure is straightforward to write once you know who’s reading it. This guide covers where the statement goes, what to put in it, and how committees, graduate schools, and journals actually look at it.

Key takeaways

• A dissertation disclosure is higher-stakes than a coursework note: it’s a permanent, published document that a committee, a graduate school, and sometimes a journal will all read.
• Check three layers of rules in order: your graduate school’s thesis or dissertation policy, your program or advisor’s expectations, and any journal or IRB requirements tied to your work.
• Disclose by task, not by tool alone. Drafting, editing, literature search, coding help, data cleaning, and translation each carry different weight in a research document.
• Put the statement where readers expect it: an acknowledgments note for light use, a methods or dedicated “Use of AI tools” subsection for anything that touched data or analysis.
• Keep your drafts, prompts, and advisor approvals. In a multi-year project, version history is your strongest evidence of authorship.

Why a dissertation raises the stakes

With a term paper, one instructor reads it once and moves on. A dissertation is different in three ways that change how you should think about disclosure.

First, it’s permanent. Your thesis goes into a repository like ProQuest or your university library, with your name on it, indexed and searchable. A disclosure you skip today is a gap someone could notice in five years.

Second, more people read it, and they read it critically. A committee, an external examiner, a graduate school formatting office, and possibly a journal editor all have a stake. Each cares about a slightly different thing, which is why one generic sentence rarely covers all of it.

Third, you have to defend it out loud. At an oral defense or viva, examiners can ask you to explain any method or claim on the spot. If an AI tool did analysis you can’t reconstruct, that’s a real problem, and no disclosure statement fixes it. Disclosure handles honesty about your process. It doesn’t replace owning the substance.

Step one: read the rules in the right order

Graduate AI policies are a patchwork, and for a thesis you’re answering to several authorities at once. Work through them in this order.

Your graduate school’s thesis and dissertation policy

Start with the formatting and submission guide, not the syllabus. Many graduate schools updated these documents to address generative AI, and some now require a specific disclosure location or even supply template wording. Search the PDF for “AI,” “generative,” and “artificial intelligence.” If your school mandates a format, use theirs over anything in this article.

Your advisor and program

Your advisor is the person who signs off and sits on your committee, so their expectation is the one that governs day to day. Ask directly and keep the answer in writing. Something like: “I’d like to use Claude to tighten the prose in my discussion chapter and to help debug my analysis scripts. The interpretation is all mine. How do you want that disclosed?” A documented yes from your advisor is worth more than any general rule.

Journals and IRBs, if they apply

If chapters are headed for publication, the target journal’s author policy matters now, not later. Many publishers follow the COPE position that AI tools can’t be listed as authors and that any use must be declared. And if your tool touched human-subjects data, your IRB protocol may need to reflect that, especially for transcription or anything involving identifiable information.

Disclose by task, not just by tool

In research, *what* the tool did matters far more than *which* tool it was. The same chatbot is trivial in one role and serious in another. Sort your use into buckets.

Worth a clear disclosure, and sometimes a methods note:

• Drafting prose that ended up in the document, even after heavy edits.
• Literature search where AI suggested sources or summarized them. You still must read and verify every source yourself, since models invent citations.
• Coding and data work like writing or debugging analysis scripts, cleaning data, or generating figures.
• Translation if you wrote in another language first.
• Transcription of interviews or audio, which can also be an IRB matter.

Usually fine to handle as light acknowledgment:

• Substantial copyediting or rephrasing of your own sentences.
• Brainstorming angles you then developed yourself.

Usually skippable:

• Spell-check and the grammar squiggle in your word processor.
• A reference manager’s formatting.

The gray zone sits between these. When you’re unsure in a permanent document, say what you did. A one-line note has never sunk a defense. A hidden tool has caused plenty of awkward meetings.

Where the statement goes, and how to word it

Match the location to what the tool touched.

Acknowledgments or a preface note (light use)

If AI only helped with writing, a short note in your acknowledgments or front matter is standard.

> “I used Claude to copyedit portions of this dissertation for clarity and to suggest smoother transitions. All research, analysis, arguments, and conclusions are my own, and I reviewed and revised every suggestion.”

A methods or “Use of AI tools” subsection (research use)

If a tool touched data, code, or analysis, a reader needs enough detail to judge its effect. Put it in methods.

> “Use of AI tools: Interview recordings were transcribed using [tool/version], and all transcripts were checked against the audio and corrected manually by the author. ChatGPT (GPT-4) was used to draft and debug Python scripts for data cleaning; the author wrote the analysis plan, verified all outputs, and is responsible for the results. No AI tool was used to interpret findings or generate conclusions.”

(Those two examples are illustrations, not official templates. Adapt the wording, and defer to your school’s format if it supplies one.)

Notice what makes them work: each names the tool and version, the exact task, and what you personally did with the output. Vague disclosures (“some AI was used”) make examiners nervous. Specific ones reassure them.

Citing AI versus disclosing it

These get tangled, and they’re different jobs.

Citing is for specific content the model produced that appears in your document, the way you’d cite any source. APA and the MLA Style Center both published generative-AI citation guidance in 2023, treating the model as the source and recording the prompt, version, and date. Use a citation when output shows up as content. Confirm the current guidance before you submit, since both styles update their advice.

Disclosing is broader. It describes how a tool shaped your process even when nothing it wrote appears verbatim. No citation format captures “I bounced my framework off a chatbot,” which is exactly why a plain-language statement exists. Many dissertations need both: a citation where you quote output, a disclosure for everything else.

A mini-scenario

Here’s how that plays out. Say a doctoral student’s education program allows AI for “language editing and coding support, not analysis or interpretation.” She uses ChatGPT to debug an R script for her regression models and to smooth the prose in her literature review. Nothing the model wrote appears as content, so she needs no citation. But she adds a methods note describing the coding help and an acknowledgments line about the editing. At her defense, an examiner asks her to interpret an interaction term, and she does it cold, because the interpretation was always hers. Disclosure plus genuine command of the work is what carries the room.

Keep the receipts across a multi-year project

A dissertation runs for years, and your draft trail is the back door to your credibility. Lock it.

• Write in a tool with version history (Google Docs, Word with AutoSave) and don’t disable it.
• Keep dated outlines, notes, and the prompts you used for any substantive task.
• Save advisor and committee emails approving specific tools.

This matters because detectors are unreliable. They report a probability, not proof, they can flag writing that’s entirely yours, and a 2023 Stanford study found they’re biased against non-native English writers. OpenAI even retired its own classifier in July 2023, citing low accuracy. If a detector misfires on a chapter you wrote, your version history and disclosure are what speak for you. For false positives, appeals, and how scores work, our other student writing guides go deeper, and if you’re weighing a checker, you can see what a checker costs first.

One more habit: before a chapter goes to your committee, read it aloud. Disclosure settles the honesty question, but it won’t make stiff, generic prose sound like a scholar’s own voice. If a passage reads like a template, it needs your thinking in it no matter where the draft started.

The honest bottom line

Disclosing AI in a thesis isn’t a confession. It’s how you stay inside your program’s rules, satisfy a journal or IRB if they’re involved, and walk into your defense with a clean account of how the work was made. Read the policies in order, ask your advisor in writing, disclose by task in plain language, and keep your drafts.

Before you hand a chapter to your committee, check how a chapter reads and make sure the writing genuinely sounds like you.