Safe AI help · Updated 19 September 2026

Using AI in a Thesis or Dissertation: Ask Your Committee First

The rule that governs a thesis is not the rule that governs your coursework. At a growing number of graduate schools, using AI anywhere in a thesis, dissertation or capstone counts as unauthorized assistance by default — prohibited unless your committee approved it in advance, in writing, for a scope you agreed with them. Disclosing it afterwards does not repair a use that was never authorized in the first place.

The default flipped, and the wording tells you so

Undergraduate AI rules mostly run on disclosure. You use a tool, you say you used it, you move on. Graduate milestone policies are being written the other way round, and the difference is not stylistic.

The University of Georgia's microbiology department states it bluntly: generative AI use in theses and dissertations is "considered unauthorized assistance per the Academic Code of Honesty and is prohibited unless specifically authorized by members of the advisory committee for use within the approved scope." Read the clause structure rather than the topic: the starting position is prohibition, authorization is the exception, and the exception comes with a boundary attached.

The University of Toronto's School of Graduate Studies says the same thing about timing: "The use of generative AI tools in any aspect of researching or writing of the thesis must be done with the prior approval of the supervisor(s) and supervisory committee." Students are told to "seek and document in writing unambiguous approval" from them "in advance of the use of generative AI tools." Three phrases do the work — in writing, unambiguous, in advance — and a supportive nod in a corridor is none of the three.

Auburn University's Graduate School requires both halves — AI use "must be disclosed and approved by the student's advisory committee" — and Illinois State extends the rule past the dissertation to "the thesis/dissertation/capstone/comprehensive exam process."

The consequence is a calendar one. The right meeting for this conversation is the proposal or prospectus meeting, when your committee is already in a room agreeing what the project is. The wrong time is the week you are formatting the final document, when the only honest options left are a disclosure that admits unauthorized use and one that is false.

Permission is one hurdle. Originality is a different one.

This catches out students who did everything right procedurally, because it is not an integrity rule at all — it is the definition of the degree.

The University of Illinois Chicago's thesis policy puts it in one sentence: "The core contribution of the thesis, as determined by the author's advisor(s) and thesis committee, must be the original work of the author and cannot be that of any other party or generative AI." Where AI has been used, UIC requires a statement as an appendix to the final document describing its role and what was authorized.

Notice what that sentence does. It does not ask whether you had permission. Permission from a committee cannot convert a machine's output into your contribution, any more than permission to hire a research assistant would. UGA's policy gives the reason in the language of the degree itself: a thesis demonstrates "independent judgment," and a dissertation represents "originality in research, independent thinking, scholarly ability, and technical mastery."

Toronto raises the same objection from the technical side, warning that AI outputs "may not meet ... the originality requirements" for a doctoral thesis because such systems "rely on existing sources to generate content-based probabilistic ... functions that may not result in sufficiently original content." Argue with that as a description of how models work if you like; it is still the view held by the graduate school that writes your regulations.

So there are two tests, and clearing one tells you nothing about the other. Was it authorized? Your committee answers that, in advance. Is the contribution yours? Nobody can authorize a yes.

The prohibited list is probably not the one you expected

Students tend to assume the forbidden thing is generated prose. Some schools are much more specific, and the specifics are surprising.

The University of Alabama in Huntsville's Graduate School publishes a list of misuses for theses, dissertations and DNP projects. Writing "significant portions of text, similar to the rules by which plagiarism is evaluated" is on it, as expected. So are these:

• "Using generative AI to gather and evaluate academic sources for literature reviews"
• "Using generative AI to form software that collects data and generates figures, graphs, and tables, and which analyzes and interprets data ... without direct involvement by the student"
• "Using generative AI to replace the creative process"
• "Using generative AI without adequate review to ensure the validity of sources"

What it permits is narrow by comparison: "formatting documents and citations" and "generating supporting visuals and illustrations that do not contain data."

The first bullet deserves a pause, because searching and triaging sources with AI is the most common student use of all — and it is a workflow we describe on this site. Our literature review workflow is written for seminar papers. At a school with UAH's wording, running the same routine on a thesis chapter is a violation even though every source in it is real and every word is yours. Nothing about the habit changed. The document it was applied to did.

Where the disclosure goes is not the same everywhere

"I will write a paragraph about it at the end" is a reasonable guess and frequently the wrong format. Four schools, four answers:

• Auburn builds it into the document — "AI use disclosure will be included as part of the official thesis/dissertation format requirements" — and publishes a sample statement with its electronic thesis and dissertation templates.
• UIC wants an appendix describing the role of AI and what was authorized.
• UGA asks that "the extent of generative AI usage should be disclosed in a statement within the thesis or dissertation."
• UAH collects it somewhere else entirely — on the Notification of Examination/Defense form — and warns that nondisclosure can end in program dismissal.

Illinois State adds the layer underneath all of them: usage "must be appropriately cited following the guidelines of the style manual used by your discipline," and "individual schools/departments may have additional policies, guidance, or restrictions." Two documents govern you, not one. Our guide to citing ChatGPT in APA, MLA and Chicago covers the citation half; only your graduate school can tell you the format half.

"There is no policy" is a temporary condition

Plenty of students check once, find nothing, and treat silence as permission for the next three years. The governance data says that is a bet against the clock.

The European University Association's Council for Doctoral Education surveyed 217 universities across 37 countries for its 2025 policy report. As reported by Times Higher Education, more than two thirds of those universities say their doctoral candidates are using AI in their research, while almost a quarter do not know the extent of it and 38% are writing AI policies for the first time.

Student-side numbers point the same way. A survey of 83 MBA thesis students in Japan, reported by Kenji Saito and colleagues in April 2026, found 95.2% reporting at least some AI use and 77.1% heavy use — one program, one country, a 36.1% response rate, so a signal rather than a global figure. But a field where use is near-universal while a third of institutions are still drafting rules is a field where the rules will move under you.

The fix is small: re-read the policy at each milestone — proposal, candidacy, defense, deposit — and keep a dated copy of the one you were told to follow. If the rule tightens between your proposal and your deposit, the written approval you already hold is what shows when you agreed to what.

Split the job

The workflow, step by step

Four prompts for ChatGPT, Claude, Gemini or whatever you already use. None of them writes your thesis, and none of them tells you what your institution permits.

1. Turn vague intentions into a scope a committee can actually approve. "Can I use AI?" invites either a nervous no or a meaningless yes. A numbered list gets you an itemized answer.

I am going to ask my thesis committee for written permission to
use AI tools in my research. I need to describe exactly what I
want to do, so they can approve or refuse each item separately.

How I expect to use AI: [describe it plainly, however it comes out]

Turn this into a numbered list of specific, separable uses. For
each one, write a single sentence naming the kind of tool, the
task, and what part of the work stays entirely mine.

Rules:
- Do not tell me whether any of these is allowed. My committee
  decides that and you cannot know their rules.
- Do not add any use I did not describe.
- Where something is vague, show me the narrowest version and
  the broadest version, so I can see what I would be asking for.
- Flag any item whose output would end up in the thesis text
  itself rather than only in my process.

Take the list to the meeting, then ask for the answer in an email — even if it is just your supervisor replying "confirming items 1, 2 and 4 as discussed."

2. Interrogate your own graduate school policy. Most of these documents answer three of the four questions that matter and leave the fourth to your department.

Below is my graduate school policy on generative AI in theses
and dissertations.

POLICY: [paste the whole thing]

Do not tell me what I am allowed to do. Answer only these four
questions, quoting the exact wording that answers each, and
write NOT STATED wherever the document is silent:

1. Do I need approval BEFORE using AI, or only disclosure after?
2. Which specific uses does it prohibit outright?
3. Where must the disclosure appear - in the document body, in
   an appendix, on a form, or somewhere else?
4. Which style manual or citation format does it require?

Then list the questions I should put to my supervisor and my
graduate school to close every NOT STATED.

Run it twice — once on the university-wide policy, once on your department's. Where they differ, assume the stricter one until someone tells you otherwise in writing.

3. Turn your log into a disclosure that is actually true. This one only works if you kept the log, which is the argument for keeping it.

Below is my own record of how I used AI while writing my thesis,
and the scope my committee approved.

APPROVED SCOPE: [paste]
WHAT I ACTUALLY DID: [paste your log - dates, tools, tasks]

Draft a factual disclosure statement in past tense, naming each
tool, what it was used for, and which chapters or sections it
touched.

Rules:
- Use only what is in my log. Do not add a tool, a task, a date
  or a safeguard I did not record.
- Do not soften anything, and do not justify anything.
- If something in my log falls outside the approved scope, do
  not write it into the statement. List it separately under
  OUTSIDE APPROVED SCOPE so I can take it to my supervisor.

If that last section comes back with entries in it, you have found the problem early, which is the only good time to find it.

4. Pressure-test the contribution. This is the originality hurdle rehearsed out loud, and it is also defense practice.

I am going to describe the original contribution of my thesis in
my own words.

MY CONTRIBUTION: [paste 150-300 words that you wrote yourself]

Do not rewrite it, do not improve it, and do not tell me whether
it is interesting.

Ask me, one at a time, the five hardest questions an examiner
could put to me about whether this contribution is genuinely
new: what existed before it, what precisely is different, what
evidence supports that difference, what a skeptic would say, and
what would have to be true for it to be wrong.

Do not answer any of them.

Answer out loud, in full sentences, without notes. Anything you cannot say is a chapter you need to go back to. Our guide to preparing for an oral exam or viva covers the rest of the defense.

Where the line is

A thesis is not an assignment that ends when it is graded. It is deposited, indexed and permanently citable under your name, which puts the cost of a fabricated citation well past a lost mark — so run the hallucination checklist and the citation audit over anything AI touched, and read every source you cite yourself. The test that settles the rest is the one your examiners apply in the room: could you explain this method, defend this claim and describe how you reached it, unaided? If not, it is not yet your contribution, whatever the permissions say.

Related reading

• The research workflow. A literature review workflow with AI and the synthesis matrix method — check your thesis policy before applying either to a chapter.
• Getting permission. Email templates for asking about AI use and the policy checklist, both of which scale up from coursework.
• Disclosure and citation. Writing an AI disclosure statement and citing AI in APA, MLA and Chicago.
• Checking the work. Evaluating sources, verifying AI answers, and reading scientific papers properly.
• The defense. Preparing for an oral exam or viva with AI, where the contribution has to be yours out loud.

FAQ

Do I have to ask permission, or is disclosing it afterwards enough?

It depends on your school, and the schools that have written this down recently mostly want both. Toronto requires approval "in advance" from the supervisor and supervisory committee, documented in writing. UGA treats unapproved use as unauthorized assistance that is "prohibited unless specifically authorized" — so a disclosure at the end would be a disclosure of a violation, not a cure for one. Auburn requires that use be "disclosed and approved." Assume you need permission first unless your own policy says otherwise in plain words.

Does this apply to a master's thesis or a capstone, or only a PhD?

Not only a PhD. Illinois State's policy covers "the thesis/dissertation/capstone/comprehensive exam process," and UAH's guidelines cover theses, dissertations and DNP projects alike. The common factor is not the degree level but the format: a supervised, committee-examined piece of independent work that gets deposited under your name. If that describes what you are writing, look for a graduate school policy rather than a course syllabus.

Can I use AI to search for and summarize sources for my literature review?

Check before you do. This is the single most common student use and some graduate schools prohibit it specifically — UAH's misuse list names "using generative AI to gather and evaluate academic sources for literature reviews" outright, and separately bans use "without adequate review to ensure the validity of sources." A workflow that is entirely normal in a seminar paper can be a violation in a thesis at the same institution.

What if my university has no AI policy for theses yet?

Treat that as unfinished rather than settled. In the 2025 EUA survey of 217 universities, 38% said they were establishing AI policies for the first time and almost a quarter did not know how much their doctoral candidates were using the tools — so a policy is likely to arrive before you deposit. Ask your supervisor for written guidance now, keep their reply, and re-check the rules at each milestone. Written approval under the old rules is far easier to explain than silence.

Where does the AI disclosure statement actually go?

Wherever your graduate school says, and the answers genuinely differ. Auburn builds it into the official thesis and dissertation format requirements and publishes a sample. UIC wants an appendix describing the role of AI and what was authorized. UGA asks for a statement within the document. UAH collects it on the Notification of Examination/Defense form instead. Ask your thesis format office, not a template you found online.

Bottom line

For coursework, the question is what you disclose. For a thesis, the question is what you were authorized to do before you did it, and the person who grants that authorization is the same person who will vouch for the work at your defense. Bring a numbered list of intended uses to your proposal meeting, get the answer in an email, keep a dated log from day one, and check where your school wants the disclosure long before you are formatting the final document. And keep the second test in view the whole way through: permission covers the tool, never the contribution.