Is Using AI for Your Thesis Cheating? Academia Is Asking the Wrong Question

Table of Contents

1. The question every workshop gets asked
2. A clear line: collaboration vs ghostwriting
3. Three safeguards: stay defensible at any time
4. 1. Before you start: check your department's rules
5. 2. While you work: verify all AI output yourself
6. 3. Throughout: the research log is your proof of thinking
7. So should you tell your supervisor?
8. In the end, what has AI changed?

Part 2 of the "work system" series. The previous post covered how to cut the barrier to learning an AI tool down to 10 minutes. This one deals with the elephant in the room: is it legitimate to use AI for a thesis?

#The question every workshop gets asked

"Professor, does this count as cheating?"

Whenever we teach graduate students to use Claude Code, this question comes up, usually within the first hour. And the people asking are rarely students looking for a loophole. They are the most conscientious ones: they are genuinely afraid of crossing a line without noticing.

The conclusion first: "Did you use AI or not?" is the wrong question.

When you run a regression in SPSS, nobody says the statistics were done by software. When you polish an English abstract with Grammarly, nobody says someone else ghostwrote the abstract. When you search the literature in Google Scholar, nobody says the literature review was written by a search engine. The tool has never been the problem.

There is only one real question: who made the judgements in the thesis?

#A clear line: collaboration vs ghostwriting

Once you lay out what AI actually does in a thesis, the line is not hard to draw.

On this side is collaboration (the tool does the labour, you make the judgements):

• Setting up the folder skeleton, tracking progress, organising meeting notes: administrative work
• Sorting 30 PDFs into folders and extracting "question / method / findings" from each into notes, after which you decide which ones are actually relevant
• Adjusting formatting to your university's template and standardising the citation style: typesetting work
• Drafting "arguments you have already thought through" into prose, which you then revise sentence by sentence: secretarial work
• Writing the code for data cleaning and statistics, with you verifying the results: no different in essence from using SPSS

On that side is ghostwriting (the judgement has been handed entirely to the model):

• "Come up with a research contribution for me," then accepting whatever comes back. The contribution is the soul of a thesis, and nobody else can think up your soul for you
• Putting AI-generated citations into your references without checking them against the originals. AI fabricates bibliographic entries that look entirely real, and at any university that constitutes academic misconduct
• Letting AI draw the conclusions in your results analysis when you lack the ability (or the will) to verify whether they are right
• Being asked in your oral defence "why did you use this model?" and having no answer, because you were not the one who made that decision

Note: the line is not drawn at "how much text the AI produced". It is drawn at whether every key decision passed through you. A thesis with a high share of AI-drafted text, where every argument and every citation has been reviewed and revised by you, is far cleaner than one typed entirely by hand whose arguments were thought up by an AI.

#Three safeguards: stay defensible at any time

Knowing where the line is is not enough. You need mechanisms that stop you from crossing it at 3 a.m. while racing a deadline. In our classes we use three:

#1. Before you start: check your department's rules

Rules on AI assistance vary widely between universities, between departments, and even between journals, and they are still changing. Spend 20 minutes before you start getting clear on the degree examination regulations, the thesis formatting guidelines, and the AI policy of the journal you are targeting. The rules outrank this article. If your department says no, the answer is no.

#2. While you work: verify all AI output yourself

Write this into your working rules (we put it directly in the project's CLAUDE.md, so the AI is reminded in every conversation):

When the AI cites literature, legal provisions or data, it also gives the source, so the claim can be checked against the original.

Then actually go and check. Especially three kinds of things: bibliographic entries (what AI fabricates most often), numbers (they drift when restated), and legal provisions and regulations (versions go out of date). Verification is not distrust of AI; it is a researcher's basic duty. When you cite data from a senior student's thesis, you still have to go back to the original source too.

#3. Throughout: the research log is your proof of thinking

This is the most underrated of the three. At the end of every work session, write four lines:

• What you did
• What decisions you made (and why)
• Where you got stuck
• The next step

Three months later, at your oral defence, a committee member asks: "Why did you change the measurement method in Chapter 3?" You open the log, and that day's entry records that you compared three options, discussed them with your professor, and why you chose the current one. This log is the hardest evidence that "the judgements were mine." (Incidentally, it is also the best source material you will have when writing the "research limitations" and "acknowledgements" sections.)

#So should you tell your supervisor?

Yes. And tell them proactively, and early.

Not out of guilt, but because it works in your favour: once your professor knows which tools you use and how, they can apply scrutiny in the right places. And when you are asked about AI use at your oral defence, "My supervisor has known about this from the start, and the method is..." puts you in a completely different position from hesitating and stumbling.

Practical advice: at your first meeting, explain your workflow clearly (what the tools do, what you do, how you verify), and ideally make your research log visible to your professor too. Transparency is the best defence.

#In the end, what has AI changed?

AI has not lowered the standard for a thesis. It has moved the weight of the standard back to where it belongs.

In the past, most of the time spent on a thesis went into "compliance labour": formatting, typesetting, finding literature, tidying the bibliography. Judgement was squeezed out late at night. Now the compliance labour can be shrunk to very little, and the remaining time goes entirely back to the questions that matter: is the question important, is the design rigorous, does the inference hold up?

In other words: in the AI era, it is harder to coast through a thesis. Everyone has the tools. Judgement is what shows.

---

This complete "collaborate, don't ghostwrite" work system (including the project template, verification rules and research log) is taught hands-on in university classrooms by UltraLab instructor beebee. His day job is financial planning (RFC / ChRP), so "every number must be verifiable" is already a professional instinct for him. Course materials are handed out in class. The next post in the series moves to a different arena: we built a 340,000-character viral-post methodology into the product, then started a 30-day field test.

FAQ

Is using AI to write a thesis cheating?

'Did you use AI or not?' is the wrong question, and the tool has never been the problem. The real question is who made the judgements in the thesis: the line is not drawn at how much text the AI produced, but at whether every key decision passed through you.

Which uses of AI in a thesis count as collaboration, and which count as ghostwriting?

Collaboration means the tool does the labour and you make the judgements: setting up the folder skeleton and tracking progress, sorting PDFs and extracting notes, adjusting formatting and citation style to your university's template, drafting arguments you have already thought through and then revising them sentence by sentence, and writing data cleaning and statistics code whose results you verify. Ghostwriting means the judgement has been handed entirely to the model: asking AI for a research contribution and accepting whatever comes back, putting AI-generated citations into your references without checking the originals, letting AI draw conclusions you lack the ability to verify, and having no answer when asked in your oral defence why you used a given model.

Can I put AI-generated references straight into my thesis?

No. AI fabricates bibliographic entries that look entirely real, and putting AI-generated citations into your references without checking them against the originals constitutes academic misconduct at any university. Write a rule into your working rules that the AI gives the source whenever it cites literature, legal provisions or data, then actually check, especially bibliographic entries, numbers, and legal provisions and regulations.

Should I tell my supervisor I am using AI for my thesis?

Yes, proactively and early. Once your professor knows which tools you use and how, they can apply scrutiny in the right places, and when you are asked about AI use at your oral defence, a supervisor who has known from the start puts you in a completely different position from hesitating and stumbling. At your first meeting, explain your workflow clearly (what the tools do, what you do, how you verify), and ideally make your research log visible to your professor too.

How do I prove at my oral defence that the judgements in my thesis were mine?

With a research log. At the end of every work session, write four lines: what you did, what decisions you made (and why), where you got stuck, and the next step. When a committee member asks why you changed your approach, that day's entry is the hardest evidence that the judgements were yours.