Ethics of studying

📋 30-second summary

• Four levels of use: explanation, organization, generation of portions, full generation. Only the first is always OK without declaring anything.
• Declare AI use in theses and papers, with a formula that says what you used it for and what you didn’t.
• University policies change: before you write your thesis, check your university’s official page.
• AI detectors are unreliable, in both directions. They aren’t the right filter, and they aren’t your filter.
• The question isn’t “what can I get away with hiding”, it’s “which work is mine”. If an oral exam or a thesis defense awaits, the rule applies on its own.

Seven lessons ago, opening the module, we set a contract: AI is a study tool, not a substitute. In seven lessons you’ve learned to use it to study a new topic, take notes, revise for an exam, write a thesis, handle sources and citations, translate, and notice when it tricks you. What’s left is the question the module has been holding back, so we could reach it with some substance in hand: where does help end and copying begin?

The answer isn’t a rigid scheme. It’s a method of judgement. The good news is that the principle is the same one from What NOT to do, in the section on work where your signature counts as your competence: whoever signs is whoever answers for it. Applied to formal study, this principle becomes a four-level scale, a practice of declaration, and a closing rule that holds even when no one is watching.

Four levels of use

Not a sharp boundary, a scale. The higher you go, the more AI substitutes your work instead of accompanying it. The first two levels are the module you’ve just read. The last two are where the line between study and shortcut gets drawn.

1. Explanation and tutor

Understanding a topic, asking clarifying questions, having a line of reasoning corrected, simulating an oral. Always OK. No declaration required: it’s equivalent to asking a coursemate, using an extra textbook, watching a lecture on YouTube. The study stays yours because the final result (having learned) sits in your head, not in a file you hand in.

It’s the pattern of Study a new topic and Revise before an exam. AI quizzes you, explains, breaks down a definition: you learn.

2. Organization and structure

A study map of your notes, the index of a paper, the chapter outline of a thesis, the order in which to present arguments. Usually OK, but if it’s in a thesis or paper, you declare it. The conceptual structure of the topic is yours: AI helps you put it into external form. The risk is low because you’re the one choosing what to put in and what to leave out, AI just gives you the container.

It’s the pattern of Notes from lectures and long texts and Write a thesis in the planning part. For personal notes nobody asks for a declaration. For a thesis, a note at the beginning or end of the document solves the problem.

3. Generation of portions of text

A paragraph, a transition between two sections, a conclusion, a rephrasing of an unclear passage. It depends on the context, and when it’s OK it has to be declared. Four cases, four different answers:

• Thesis. Usually forbidden, or allowed only with an analytical declaration (some universities ask you to specify which sections, which prompts, which model version). See the callout on institutional policies below.
• Academic paper. The paper has to be yours in its words. AI as a style reviewer (it rereads you, flags awkward sentences, suggests synonyms) is generally OK. AI as the author of paragraphs isn’t: that text wasn’t written by you. The practical line: if you decide the substitution word by word, it’s revision; if you accept AI’s text in blocks and polish it, it’s already generating.
• Graded essay. Plagiarism, full stop. If the text counts as your assessed written output, generating pieces and passing them off as yours is copying.
• Ungraded work (a post for the course blog, a brief, an internal email to a study group). The rule of work applies, not the rule of study. Module 3 For work covers it.

4. Full generation

AI writes the text, you correct it and rephrase a bit. In formal assessment this is copying, regardless of how much you clean up the output. A competent examiner can hear the “thread”: a fluency that doesn’t belong to you, a lexical choice off-register, an argument you can’t defend when they ask “why did you write this here?”.

It’s not a matter of risk (AI detectors aren’t reliable, we’ll come back to it). It’s a matter of substance: you haven’t learned what the professor wanted you to learn. The assessment, even if it goes well, evaluates something that isn’t yours. It holds even if you rewrite the generated draft word by word: the argumentative structure stays AI’s, and that’s the part the professor wanted to see you build.

Declaration of use, how it’s done

More and more universities require a short note at the beginning or end of theses and papers, in which the writer declares what they did with AI and what they didn’t. When the request exists, the exact formula is in your university’s guidelines. When it doesn’t, declaring anyway is good practice: it protects you (in case of questions at the defense) and makes the work verifiable.

A template formula, adaptable to your case:

In the writing of this work, ChatGPT (version […], model […]) and Claude (version […], model […]) were used as:

1. study support (explanations, simulated quizzing, revision);
2. second readers for style revision, clarity suggestions, simulation of objections from the supervisor;
3. formatting of bibliographic citations (manually verified through DOI).

Generative AI was NOT used for:

1. generating finished paragraphs or conclusions;
2. producing the bibliography (sources were found and verified via Scopus/JSTOR/Google Scholar);
3. translating or rephrasing passages without the author’s intervention.

Always check your university’s instructions for the required formula: some universities ask for detail per section, others just a general note. If there’s no imposed formula, the version above is a good starting point. The [...] placeholders go filled in with what you see in the product’s interface at the time of use (e.g. ChatGPT, Plus plan, GPT-5 model on 12 May 2026; Claude Sonnet 4.6, web app on 12 May 2026). If you used the product across different months and the models changed, you declare it.

Institutional policies change

Three examples cited as a pattern, not as exact sources. Politecnico di Milano has published guidelines on the use of AI in graduation work. MIT runs a generative AI consortium with guidance for instructors and students. Stanford University has issued official guidance on AI in coursework and an academic integrity policy with explicit references to generative AI. Official pages change often, so I don’t link specific URLs: before writing your thesis, look up the policy on your university’s site (usually under “research ethics”, “code of conduct”, or “graduation work regulations”). If you can’t find an explicit policy, the general rule applies: use AI as a study tool, declare what it did, don’t let AI write what you sign.

Proctored exams and AI detectors

During a supervised exam, AI is forbidden. Even if the exam is at-home with webcam (proctoring) and even if the university doesn’t technically block access to ChatGPT or Claude. It holds for written exams, for online simulated oral exams, for timed tests. The rule isn’t “if you don’t get caught it’s OK”: the rule is “if they ask you to do it on your own, you do it on your own”.

AI detectors (Turnitin AI, GPTZero, Copyleaks and similar) exist, and some universities use them. They’re unreliable. Turnitin itself, in its official disclaimer, writes that the detector’s score shouldn’t be the only basis for an academic integrity decision: they produce false positives on human texts with a regular style (people who write cleanly and in a structured way look like an AI) and false negatives on AI text heavily rewritten by hand.

The closing rule, picked up from What NOT to do: the bar isn’t “what can I get away with hiding”, it’s “which work is mine”. With that filter, AI detectors and concealment strategies become irrelevant: the question is internal, not procedural.

The “whoever signs” rule

I’m adapting to the study context the rule from What NOT to do, in the section on work where your signature counts as your own competence. Three direct applications:

• You take the oral. If AI did the work for you, you fall apart at the first exchange. The professor asks you to argue a point from your paper, you don’t know what you actually wrote, and the conversation collapses in two minutes.
• You defend the thesis. If AI wrote it for you, your supervisor notices at the defense. A direct question about the third chapter, you can’t answer, and the committee understands you didn’t do the work.
• You pass the exam. If you studied a dumbed-down version from AI instead of the textbook, you show up with a preparation the professor fails. You invested time, but in the wrong place.

What NOT to do

• Don’t have AI write what you’ll sign (exam, thesis, paper). The signature is your declaration that the work is yours. If it isn’t, the signature is false.
• Don’t use AI detectors to “optimize” AI output so it gets past the detector. It’s an unethical use (you’re trying to fool the system) and unstable (every detector update can unmask a text that used to pass).
• Don’t think “no one will know”. People who correct theses and papers for a living see the pattern: strangely smooth prose compared to the notes you sent them, perfect transitions where there are usually jumps, conclusions that read like a Wikipedia summary. They often know, and they always suspect.

Check what you’ve understood

📝 Check-in with your AI

Ethics of studying in five steps

5-7 min of check-in

Paste the text into your AI. It'll ask five progressive questions on the four levels of use, on declaration in theses and papers, and on the whoever-signs rule.

Copy the text below and paste it into your favourite AI (Claude, ChatGPT, Gemini, or another). Or download the file and give it to your AI as an attachment.

Copy the text Download Open

What comes next

AI is a real study tool, used well. It’s a destructive shortcut, used badly. The eight lessons of this module have shown the first use: how to explain a new topic to yourself, how to organize notes, how to revise, how to approach a thesis with method, how to handle sources and citations, how to translate and compare texts, how to notice when AI tricks you, and how to stay on the right side of the line when the work needs to be signed.

From here on the manual opens onto other audiences. For those who teach, Module 5 For teachers takes the same questions from this module and flips them from the lectern: how to design assessments that hold up against AI, how to ask students to declare their use, how to recognize the pattern of generated text. For those who want to understand how AI does what it does, all the way down to the technical level, Module 6 For those who want more starts here and goes under the hood. For now the student module is complete: come back to these lessons when you need them.