Thesis Writing with AI: A Complete and Honest Guide
If you are writing a thesis or a dissertation today, you have probably turned to an AI tool at least once: to find an angle, to organise your thinking, or simply to make a paragraph read better. The real question is not whether these tools should be used at all. The real question is where the boundary sits, and which part of the work has to stay entirely yours.
This guide is honest in both directions. It is not an advertisement for AI tools, and it is not a scare story about them. It starts with what AI actually does in thesis writing, then separates the three layers of research work, lists the red lines, and closes with a practical checklist. If your main question is whether using AI amounts to plagiarism, a companion article in this cluster answers exactly that.
What AI actually does in thesis writing
The genuine uses of these tools are wider than most people assume and narrower than the marketing suggests. Brainstorming to find an angle on a problem, turning a rough idea into sharper research questions, and arranging the map of your chapters are all tasks where a tool is fast and useful. Summarising an article you have already read, translating text you wrote yourself, editing your language and unifying your tone, and getting help writing the analysis code for your own data belong to the same group.
The subtle point is that every one of these is instrumental help. When help is instrumental, you are still the decision maker and the one who understands: the tool tidies your text, but the content of that text came from your reading and your thinking. Replacing the author happens at a different moment, when the tool thinks in your place and writes in your place. At that point the final text is no longer yours, however fluent and polished every sentence looks. Everything that follows in this article rests on that one distinction.
The three layers of thesis work and where AI stops
The first layer is thought: defining the problem, formulating the research question, choosing the method, and interpreting the findings. This layer cannot be delegated. The reason is not only ethical but technical. Until you have built the problem yourself, you cannot defend it, and you do not know why you chose this method rather than another. Any tool that enters this layer is in fact doing the central work of the research.
The second layer is execution and analysis: writing the analysis code, cleaning raw data, building tables, and doing repetitive work. Here, taking help from a tool is reasonable, on one condition: transparency. That means knowing exactly what the code or the calculation does, being able to explain it line by line, and, where appropriate, noting in the text or an appendix that a tool assisted with this part. If you cannot explain it, you have not understood it, and you will be caught out in the defence.
The third layer is language and editing, and it is the safest of the three: unifying tone, fixing Persian half-spaces, shortening long sentences, and rewriting your own text. Even this layer has a condition, though. The input has to be your own text, not somebody else’s and not text that was generated in your place. If the input is another person’s work, academic misconduct can enter through the door marked language.
What you should never hand over to AI
There are five specific tasks where delegation goes beyond a small mistake and becomes misconduct. These are the clear red lines of this discussion:
- Fabricating data, or filling in questionnaires and interviews instead of collecting real data;
- Inventing sources and references, or citing a paper you have never read;
- Having the tool write the discussion and conclusion where the reading of the data has to be yours;
- Delegating methodological decisions, from sampling to the choice of analysis method;
- Collecting data from participants through a tool without their knowledge.
The most dangerous of these is the invented reference, because it is the fastest route to misconduct. Citing a work that does not exist, or that you have never read, breaks research ethics and is also easy to expose in review and textual similarity checks. A healthy rule is simple: every source that appears in your reference list should have been open on your desk.
The right path from proposal to defence
At the proposal stage, a tool can help you tidy the structure of your plan, separate your objectives from one another, and write the research question more clearly. The definition of the problem and the research gap, however, have to be your own work, because the whole thesis and the whole defence rest on those two. If you let the tool generate content at this stage, you will be stuck in the later chapters.
At the writing stage, order matters: write the chapter yourself first, then call on the tool for editing and polishing. The same proportion holds at the analysis stage. Help with writing code is acceptable, but the decision about what to measure and how to interpret the result is yours. On registration with IranDoc and the similarity check before the defence, remember that this step is a mandatory procedure in postgraduate study, and that the Hamandjoo system shows the textual similarity between your work and other dissertations, theses, articles and web text.
At the defence stage, prepare one thing in advance: an honest answer to the question of what role a tool played in your work. If you have been transparent from the start, that question is not a threat; it is a chance to show that you own the work. Keep the trail as well. Drafts show how the text was built.
If your supervisor or university is strict
Strictness here is not strange, because university regulations are not uniform on this subject. The four-step rule below carries the least risk: ask, get permission, disclose, and keep it in writing. First ask your supervisor or the research office what practice is accepted at your university. Then, if you hand in a permission request, get an answer. Then disclose the role of the tool in the text or the preface. And keep written replies at every stage. Those four steps protect you from improvised, verbal decisions.
Know one fact clearly as well. The governing text on academic misconduct in Iran is the Law on Preventing and Combating Fraud in the Preparation of Academic Works, passed by the parliament, together with its executive regulation approved by the Council of Ministers. Neither the law nor its regulation contains any clause about text produced with artificial intelligence. In other words, there is no explicit ruling on this subject today, and the decision ultimately falls back on university regulations, defence practice, and your own disclosure and citation.
If your university uses the output of AI text detectors, note too that such output alone is not definitive proof of misconduct. These tools are error-prone. The documented example is that the maker of one of the best known of them retired the classifier that detected its own text, announcing low accuracy. The practical consequence is that false positives exist, and fully human writing can be flagged as well. So put the standard for judgement on transparency and correct citation, not on playing games with these tools.
Skills that must never be delegated
A few skills are the backbone of a researcher, and no tool can replace the practice they require: formulating a research question, designing a method that genuinely answers that question, reading sources critically, writing analytically, and editing your own text deeply. These are exactly the things measured in a defence as well.
Internationally, publisher practice is clear: the use of an AI tool must be disclosed at the time of submission, and a tool cannot be the author of an article. Responsibility for the accuracy of the text and for avoiding plagiarism rests entirely with the human author. So even where disclosure is accepted, the replacement of thought is not.
Conclusion and a practical checklist
The Ministry of Science has circulated a guide on using AI tools in research, and according to media reports the output of AI text detectors alone cannot be a definitive basis for proving misconduct. The conclusion of this article is the same. Use tools in the instrumental role, build the thinking and the data yourself, and keep your working trail transparent. If you want the precise cases of acceptable use, this cluster has two related pieces: legitimate ways to use AI in academic writing and is using AI in a thesis plagiarism or not?. To build the framework of your argument, from idea to scientific framework helps, and if you want to review the overall shape of a scientific article, writing a scientific paper and the IMRAD structure is a good starting point.
Run through this short checklist once before you hand in any chapter:
- I wrote the problem and the research question myself and can explain them in a minute.
- I chose the method and the analysis myself and I understand the logic behind them.
- Every figure and value in the text goes back to my own real data.
- I have actually read every source I cite.
- The input to every editing tool was my own text.
- I have disclosed the role of any tool in the right place.
- I have kept the drafts and notes that show the working trail.
None of these seven items is difficult; they are all just discipline. Keep that discipline and you get the real benefits of AI while still defending your work with confidence.
