Legitimate Ways to Use AI in Academic Writing: Is There a Way Forward?

If your conscience keeps saying that you should not use AI at all, while you can see that a tool would remove hours of repetitive work, the real question is this: is there a legitimate way forward? The short answer is yes, provided the tool is given an instrumental role rather than a thinking one. Instrumental means arranging, polishing, translating and writing code; thinking means deciding on the problem, the method and the meaning.

In practice the permitted line is clearer than people assume, and it rests less on an ethical rule than on a rule of proportion: the greater the share of thinking or data that a tool produces, the further the work moves beyond the boundary of help. This article lists seven acceptable tasks, shows where ordinary services sit in the law, explains the right order of work, and offers a sample disclosure statement.

Seven tasks that are acceptable

These seven tasks take a large part of the daily writing load off your shoulders while leaving the share of thinking untouched:

• Brainstorming and question building: producing a list of possible angles on a problem.
• Building a chapter map: arranging the order of chapters and subsections before you write.
• Language rewriting of your own text: making sentences flow without changing the meaning.
• Summarising articles you have already read, for your own notes and review.
• Building a comparison table from sources you have studied yourself.
• Translating your own text, from Persian into English or the other way round.
• Help with analysis code on your own data: writing or fixing code for repetitive calculations.

What these seven have in common is one thing: the input is always yours. You read the article, you wrote the text, you collected the data; the tool only improves the shape of the work. With that single condition in place, you do not have to argue with yourself every time about whether a particular task is permitted.

Where ordinary services sit in the law

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. The law identifies the offence specifically where a work such as a dissertation, thesis, article or research project is prepared, offered or handed over for profit and as a profession, with the aim that another person presents all or part of it as their own work. The focus of the law, in other words, is the handing over of a work for profit, not the use of a tool.

More important still, ordinary services, meaning typing, translation, editing, help with data collection and laboratory services, fall outside the scope of that law. An instrument of practical help is therefore not an offence in itself. At the same time, neither the law nor its regulation contains any clause about text produced with artificial intelligence. There is no explicit ruling on AI today, and the decision falls back on university regulations, defence practice, and your own disclosure and citation. The real boundary is not where a tool is used; it is where you replace thinking or data.

The write-first, tool-second pattern

The order of work matters more than the work itself. For every chapter, write a rough version with your own hand, even if it is incomplete, and only then give that version to a tool for editing and tidying. The difference between this rough version and a text a tool generates from nothing comes down to two things: your voice is audible in the text, and you know where every sentence came from.

The same order works in analysis: first decide yourself what you are going to measure, then get help writing the code, and finally interpret the result yourself. The value of the pattern becomes obvious in the defence, where you have to explain in a few seconds why a particular table looks the way it does. Someone who built and understood the text has that explanation. Someone who ordered the text does not.

How disclosure is written

Disclosure is not complicated. A short paragraph in the preface or an appendix is enough. In it, say which parts of the work received help from a tool, what the tool did, and that the content, analysis and interpretation are your own. This is the cautious practice, but it does not replace asking your own university; if your university has a defined procedure, follow that. Internationally, publisher practice is the same: the use of an AI tool must be disclosed at the time of submission and a tool cannot be the author of an article, with full responsibility for the accuracy of the text resting on the human author.

A simple sample you can adapt in your own words:

In writing this thesis, AI tools were used for language editing, for arranging the structure of the chapters, and for writing part of the analysis code. The choice of problem, the design of the method, the data collection, the interpretation of the findings and the conclusions are the work of the author, and responsibility for the accuracy of every part of the text rests with the author.

The wording above is an example, not a required sentence. What matters is that it is precise, brief and unambiguous.

Keep the trail of your work

Drafts, the notes recording the questions you put to a tool, saved versions and dates are all evidence of how your work progressed. If a question ever arises, that trail shows how the text was built and which parts moved forward with your own thinking. So do not delete versions, and keep the notes somewhere you can find them.

The same trail helps at the similarity check stage. The Hamandjoo system shows the textual similarity between your work and dissertations, theses, articles and web text, and it has nothing to say about who wrote the text. Registration with IranDoc and the similarity check before the defence is a mandatory procedure in postgraduate study. When you hold the intermediate versions of your work, answering the outcome of that check becomes much easier.

Red lines

Against the list of acceptable tasks, a few actions have no justification at all: fabricating data, inventing sources or references, not collecting real data, delegating methodological decisions to a tool, and failing to disclose the role of a tool where university regulations require it.

There is another red line that is noticed less often: hiding the use, in the hope that nobody will find out. Even when the text itself is sound, concealment damages your credibility in review. The reliable route is transparency. Disclosure never costs as much as being caught hiding something.

If they ask in the defence: did you write this yourself?

An honest answer is the best answer, and preparing it is not hard. For instance: I wrote the text myself and used a tool for language editing and for arranging the chapters; I have also described its role in the preface. That sentence does not put you in a weak position, because there is nothing to hide.

If the question turns to AI text detectors, know that their output alone is not a definitive basis for proving misconduct. These tools are error-prone, and fully human writing can be flagged as suspicious too. The documented example of that unreliability is the decision of the maker of one of the best known of these tools, which retired the classifier detecting its own text while announcing low accuracy. So instead of arguing about detector accuracy, talk about the working trail and the citations.

Conclusion and practical work

The conclusion is short. Give the tool an instrumental role, keep the thinking and the data yourself, follow the write-first, tool-second order, and disclose its role clearly. For a closer definition of plagiarism and its different cases, read is using AI in a thesis plagiarism or not?, and to review the overall shape of a scientific article, writing a scientific paper and the IMRAD structure is a good starting point.

Add one small practical habit as well. Before you hand in any chapter, write down the map of contributions: which decision was mine, which part a tool executed, and where I disclosed that. If you want your text criticised impartially before you meet your supervisor, look at the Nevispar AI review panel in this cluster, which assesses quality, structure and the strength of the argument without replacing human review.