Using AI in Your Thesis: Plagiarism or Not?
The question almost every graduate student asks at least once is this: if I use AI, does my work count as plagiarism? The honest short answer is that today there is no explicit legal ruling for text produced with artificial intelligence; the decision falls back on your university regulations, on defence practice, and on your own transparency. But having no explicit ruling is not the same as being free to do anything.
This article separates three different cases: editorial and language help, content generated by a tool, and fabricated data or sources. These three do not carry the same ethical or regulatory verdict, and they depend above all on two things: disclosure and citation. If you want to know where the line of legitimate use actually sits, legitimate ways to use AI in academic writing continues the discussion.
What plagiarism actually is
Plagiarism means taking another person's idea, text, data or result and presenting it as your own without clearly naming the source. Fabrication and fraud belong to the same family: inventing data that does not exist, manipulating a result, or reporting work that was never done. The difference between plagiarism and inspiration comes down to one thing: when you are inspired you understand someone else's idea, extend it and cite it; when you plagiarise you replace it and call it yours.
In Iran, the governing text for these violations is the "Law on Preventing and Countering Fraud in the Preparation of Academic Works" (passed by Parliament) together with its executive regulation (passed by the Council of Ministers). The most important point to state plainly is this: neither the law nor its executive regulation contains any article about text produced with artificial intelligence. No explicit ruling for AI has been written, so the decision falls back on university regulations, defence practice, and disclosure and citation.
A second point in the same law matters here: it treats preparing, offering or handing over a work (dissertation, thesis, paper, research project) for profit and as a profession, with the aim that another person presents all or part of it as their own, as a crime. Alongside that, ordinary services — typing, translation, editing, help with data collection, laboratory services — fall outside the scope of that same law. In other words, an executive aid is not in itself a crime.
Three cases you must not mix up
Case one is editorial and language help. You wrote the text yourself and the tool only makes it smoother, evens out the tone, or turns your Persian sentence into English. This sits inside the scope of those ordinary services, which the law does not treat as a crime, and in practice it is much like hiring an editor. The condition is that the content and the thinking come from you, and that you state the tool's role somewhere clearly.
Case two is content generated by a tool without disclosure. Here the tool formulates the problem, builds the argument or writes the chapter from scratch, and you submit it without naming the tool's role. The problem in this case is not the use of a tool; the problem is hiding its share and presenting the text as your own understanding and analysis. To a reviewer this can amount to a lack of academic honesty, and the final decision about it rests with your university regulations.
Case three is fabricated data or sources. Inventing a number that was never measured, a result with no test behind it, or a citation to a paper that does not exist is always a violation, and no regulation or practice permits it. This is the most dangerous case of the three, because it surfaces quickly at a defence and its consequences are far heavier than a comment about your language.
Why AI detection is a poor standard of judgment
AI text detectors are error-prone. The clearest documented example is that the maker of one of the most famous of these tools retired its own text classifier while announcing low accuracy. The practical consequence for you is straightforward: false positives exist, and a completely human text can be flagged as AI. Nobody should therefore read a single tool report as equal to the claim that a tool wrote the text.
The second point concerns procedure. The Ministry of Science has issued a "Guide to Using AI Tools in Research"; its details should be read through media and news agency reports, not treated as a definitive text. The important point those reports have relayed, and one worth stressing, is this: the output of AI text detection tools alone cannot be the definitive basis for proving a violation. If such a report ever lands on a reviewer's desk, it is an indication, not a verdict.
Add to that the fact that these tools do not measure quality either. An ordinary human text may slip past the filter, while a tool-assisted text that is tidy and polished may not be flagged at all. Building an ethical decision on that kind of instrument is needlessly risky for the reviewer and the student alike.
What a similarity check shows, and what it does not
Irandoc's "Hamandjoo" system compares your text with the text of theses, dissertations and papers as well as with the web, and shows textual similarity: the matching passages, the degree of similarity and the source of each match. The report tells you which parts of your text overlap with other texts and which source lies behind each overlap. Registering with Irandoc and running the similarity check before the defence is a mandatory procedure of graduate study.
But one thing it does not tell you is who wrote the text. Hamandjoo does not measure whether the author was a human or a tool. So it is wrong both to treat the system as a way of identifying the author and to say that "a low similarity score means my work is clean". Similarity checking and AI detection ask two different questions, and neither is a scholarly judgment on its own.
What protects you: transparency, citation, a paper trail
Three habits protect you in practice. First, transparency: saying yourself, without being asked, which parts used which tool and for what purpose. Second, proper citation: every idea, dataset and quoted sentence has a clear source, and no citation points to something you have not read. Third, keeping your paper trail: intermediate drafts, notes on your questions, the path of your analysis, and dated versions.
The paper trail matters most at the defence. When a reviewer asks how a chapter took shape, the file that shows the path of your thinking is stronger than any spoken explanation. To see what these three habits look like as a daily routine, read the rules you should follow when writing a thesis, where the same points are turned into a working checklist.
Real consequences, without exaggeration
The consequences of academic misconduct are mostly administrative and reputational rather than criminal: the loss of scholarly credibility, the work being sent back for correction or rewriting, and university measures based on that institution's own regulations. The scope and severity of those measures differ between universities and depend on how serious the violation is, on your record, and on the reviewers' judgment. Anyone who sketches a certain outcome for you is simplifying.
What makes all of these outcomes expensive is the time and credibility they cost: months of work return to zero, and the trust of your supervisor is damaged as well. That is why the most cautious route is also, in practice, the cheapest one: keep the tool's role small and clear, and carry the intellectual weight of the work yourself.
Conclusion: the three-question decision rule
Before you use a tool, ask yourself three questions. First: if someone asks me tomorrow how this part was made, can I answer without embarrassment? Second: does this part rest on my own understanding or data, or has it taken the place of my thinking? Third: are the tool's role and the sources of the ideas clear in the text? If all three answers are clear, you are working inside safe ground.
Today's law has no explicit ruling about artificial intelligence, and that empty space puts the responsibility on the writer. Transparency and citation are what defend you in the absence of an explicit rule. To see the whole picture, start from writing a scientific paper and the IMRAD structure.
