How to declare AI use in your thesis (with example statements)
It's a few weeks before you submit. You open the thesis submission form and find a new section on generative AI. Do you tick "no AI used" or "AI used"? And if you tick the second, what do you write underneath?
You've used AI on and off for three years. It suggested search terms for your literature review, tidied a few clumsy paragraphs, and once helped you fix an error in your code. Does all of that count? Some of it? None of it?
Many doctoral researchers get stuck here, and many worry that admitting to AI use will make their examiners suspicious. A clear, accurate declaration does the opposite. It's a chance to show you were in charge of your work. The real risk is AI use you never mentioned coming out later. Picture your viva. An examiner asks why your literature review opens with one debate before another, and you can't say, because AI chose the order and you never questioned it.
Here's how to write a declaration you can stand behind.
Start with your institution's rules
Your university's rules come first. Where anything in this post differs from what your institution asks, follow your institution.
You'll usually find the rules in one of four places: your doctoral college or graduate school pages, the thesis submission regulations, the submission form itself, or your supervisor. Read them again just before you submit. AI policies change quickly, and the version you read in your first year may have been updated since. The rules can also differ for your upgrade or confirmation report, your final thesis and any journal articles, so check the ones for the document in front of you. If something is unclear, ask your supervisor or your research degrees team, and keep their answer in writing.
How two universities handle it
Universities handle AI declarations in very different ways. Here are two randomly chosen examples, as their guidance stood in September 2026.
King's College London gives you two options to choose from. When you submit your thesis, you pick one of two declarations on the submission form. The first says no part of the thesis was generated by AI, and you can still tick it if you used software to check your spelling and grammar. The second says you used AI software "to rewrite, rephrase and/or paraphrase parts of this thesis to ensure the quality and standard of the English used", and asks you to name the tools you used and link to them (King's College London guidance). That second option covers help with your English. If you used AI for anything else, such as your analysis, ask your supervisory team how to describe it.
The University of Georgia in the US asks for permission first. Its Graduate School treats AI use in a thesis or dissertation as unauthorised help unless your advisory committee has approved it, and you can only use AI for what they approved. If you have that approval, you include a statement in your thesis describing "the extent of generative AI usage" (University of Georgia Graduate School policy). Georgia gives no template, so the wording is up to you.
Different universities handle AI differently, so make sure you know what your university asks of you; the rest of this post will help you write a declaration that covers everything it should.
What needs declaring, and what doesn't
Some AI use needs a full description in your declaration, and some needs none at all. In our book, Michael Rowe and I sort it into three kinds.
Everyday help with spelling, grammar, quick look-ups and formatting. Most universities count this as ordinary editing, so you usually don't need to mention it. The King's College London form says as much. If your university asks you to declare every use, including this one, then declare it.
Handing AI a task, such as summarising a paper or drafting a phrase you then rewrote. Mention it if it affected your work.
Use that shaped your argument, your evidence or your analysis. Your declaration needs to describe this in detail.
One more check to make early: many universities restrict or ban using AI to create or change images or data. Find out before you start using AI for either.
The four parts of a good AI declaration
A complete declaration has four parts:
- The tool and its version. Different versions of the same tool can give different results, so your reader needs to know exactly what you used. Add the month and year too.
- What it did. Name the task: flagging unclear sentences, suggesting search terms, helping to write code.
- Where it fits in your process. Which chapters, and at what stage: planning, drafting, analysis or editing.
- A statement that the final work is yours. This is the most important part. It says you made the decisions and take responsibility for them.
Where your declaration goes
It depends on what the AI did.
If AI helped with your writing, put your declaration in the acknowledgements, on a declaration page, or on your university's submission form.
If AI helped with your analysis, for example by cleaning data, writing code, or setting up or applying a coding scheme, describe that in your methods section. Add one more detail here: how you checked what the tool produced. If the tool carried out part of your method, you cite it too: see How to cite ChatGPT in your thesis.
Keep your supervisor in the loop throughout. Tell them as you go what you worked out yourself, where AI helped, and what you kept or threw away. Your final declaration then sums up conversations you've already had, and nothing in it will surprise them.
Example AI declaration statements
Change the details in these examples until they describe exactly what you did. Replace the square brackets with the name of your tool, its version, and the month and year you used it. Remember to check your university's guidance before you use AI. Just because an example is listed below does not mean your university is allowing you to use AI in this way.
Help with language while drafting (acknowledgements):
[Tool name and version, month and year] was used while I drafted Chapters 3 and 5 to point out unclear sentences and suggest other ways of phrasing them. I accepted, changed or rejected each suggestion myself, and the argument, structure and final wording are my own.
Early literature scoping (acknowledgements, or methods if your field expects it):
During the scoping stage of my literature review, I used [tool name and version, month and year] to identify alternative terms for my key concepts, including terms used in neighbouring disciplines. I built the final reading list from documented searches of [databases], I have read every source cited in this thesis, and the selection and interpretation of the literature are my own.
Help with analysis code (methods section):
I used [tool name and version, month and year] to help write the code for the regression models reported in Chapter 4. I checked each line against the package documentation and tested the code on a dataset with known results before running it on my own data. All analytical decisions, including the choice of model and the handling of missing data, were mine, and I can explain each step of the analysis.
Here's one to avoid:
AI tools were used in the preparation of this thesis.
It's true, and it tells your reader nothing. Which tool? What did it do? Which chapters? Who is responsible for the result? It has none of the four parts, so your examiners will have to ask you, and they may well ask in the viva.
Acknowledging AI and citing AI
Declaring and citing are easy to mix up. A citation sends your reader to a source they can find, read and judge for themselves. AI output changes between sessions and versions, and a shared link only takes your reader to the whole conversation. So you acknowledge the help AI gave you, and you cite the studies it helped you find, once you've read them.
There are two exceptions. If the AI's output is what you're analysing, for example when you discuss a response it gave or show a made-up reference as an example, you cite it. You also cite the tool when it carried out part of your method, such as sorting survey responses into categories, and you describe that in your methods section. I explain how to do both, with examples, in How to cite ChatGPT in your thesis.
Describe what actually happened
"I drafted this section and used AI to check grammar and clarity" and "I used AI to generate a first draft, which I then revised" describe two very different ways of working. In the first, you did the thinking from the start. In the second, AI did the first round of thinking, and you refined it.
When you've written your declaration, read it back. If describing what you did makes you uncomfortable, it may mean you've leaned on AI more than you meant to. To avoid this situation in future, write your declaration before you start using AI in your thesis, as a way to declare "this is how I'm going to use AI". If it makes you feel uncomfortable, you still have a chance to modify how you work with AI so that the thinking remains yours and you remain in the driving seat.
What your declaration can't do for you
In our book, we tell the story of Hannah, a PhD student in social history. Her department allows AI as long as she discloses what she used and where. She has followed every rule. Her disclosure form is complete, and every reference leads to a real article she has checked. Then she rereads her literature review chapter and realises she can't explain why it's built the way it is: why it opens with one debate rather than another, or why one question gets three paragraphs and another gets one. Her declaration is perfect. She still can't defend the chapter.
Your declaration shows your examiners that you followed the rules. In the viva, they'll want to know something else: why you made the choices you did. Why this question? Why this method? Why did you leave that study out? Only you can answer those.
The easiest way to write a strong declaration, and to answer those questions, is to keep a short record as you go. Each time AI shapes a decision, write a line or two: what it suggested, what you decided and why. For example:
14 May, discussion chapter. Asked AI to argue against my main interpretation. It raised participant age as a possible confounder, which I hadn't addressed. Checked two papers on this, added a paragraph to the limitations, and kept my interpretation. The reasoning and wording are mine.
After a few months of entries like that, most of your declaration is already written, and you have a record of your reasoning to revise from before the viva.
Want to go further?
Our book, Still Yours: A Doctoral Researcher's Guide to AI, goes into the work behind the declaration. Michael Rowe and I explain which uses of AI need a record, how to keep that record light enough that you'll actually keep it, and how to use it to prepare for your viva. We also give you a simple test for checking whether a paragraph is really yours before it goes anywhere near a declaration form.
The book is coming soon from Springer Nature.
The views in this post are my own and don't necessarily reflect those of my employers.
0 comments
Leave a comment
Please log in or register to post a comment