Academic Integrity and AI: A Student's Guide

Academic integrity AI: the phrase covers a spectrum wide enough to include running a finished paragraph through a grammar checker and asking a chatbot to write the paragraph in the first place, and most students never get a straight answer about where one use ends and the other begins. That line is not drawn once, centrally, for every course a student will ever take. It is drawn separately by each instructor, in each course handbook, and the version that governs a specific assignment is very often narrower than whatever the university's general policy says.

That isn't a loophole. It is how nearly every institution that has actually written down a policy has chosen to handle this, because a lab report, a personal essay and a coding assignment raise different questions that one central rule cannot answer well. The guide sets out what tends to be fine everywhere, what tends to be fine nowhere, the grey area in between, and a simple way to keep a record of what was actually done. So the answer to a specific question is never just a guess.

Academic Integrity AI Starts With Your Course Policy, Not Your University's

Universities that have published detailed AI guidance consistently push the actual decision down to a smaller unit than 'the university' as a whole. Oxford's policy states plainly that for students, use of AI in assessments is only allowed when explicitly permitted and must be declared following department or faculty instructions, and that unauthorised use is considered academic misconduct by default. Cambridge sets the operative rule at the level of the individual assessment brief: using unacknowledged AI-generated content in a summative assessment counts as misconduct unless explicitly stated otherwise in that brief, which means the same module can permit one kind of use in one assignment and prohibit it in the next.

MIT Sloan states this even more directly, saying it has no definitive policy on using generative AI tools to complete assignments, and that faculty set their own rules course by course, including permitting or prohibiting some or all such use. Monash devolves the decision to the Chief Examiner of each unit, who must specify where and how AI may be used, with acknowledgement always required when it is. None of these are edge cases. They are how four different universities on three continents have each chosen to structure the same policy question, which is worth remembering before assuming a school's general statement is the whole story. A closer look at university AI policy across institutions shows the same devolved pattern almost everywhere it has been checked.

The Uses That Are Almost Always Fine

The clearest pattern shows up outside universities entirely, in how academic publishers treat the same question for authors submitting finished research. Elsevier's policy states that basic checks of grammar, spelling and punctuation do not need a declaration at all. Wiley excludes tools used solely for spelling, grammar and general editing from its disclosure requirements. IEEE describes grammar and editing tools as generally outside the intent of its AI policy in the first place. None of that is a university's plagiarism rule, but the underlying distinction, a mechanical correction to writing that is already a student's own versus new content that was not, is the same one almost every course-level AI policy draws too. A free grammar checker sits squarely inside that fine category under nearly any policy a student is likely to encounter.

This works the same with feedback on structure. Feedback on existing content doesn't have to be a policy statement, sentence-by-sentence. A tool can ask you if an argument you wrote is logical, or if you repeat yourself somewhere. This isn't asking for new content to replace old content. The distinction that keeps recurring across every policy above is authorship of the substance, not which tools touched the document on the way to a final draft.

The Uses That Are Almost Never Fine

Generated argument sits on the other side of that same line almost everywhere it has been written down: asking a chatbot for the actual claim, the actual analysis or the actual interpretation, and keeping its output as the substance of the submission. Generated evidence is worse, not better, because a fabricated statistic or a study that does not exist is not a style problem. It is a false claim submitted as fact. Citations carry the same risk in a more specific form: language models regularly invent sources that read as entirely plausible, which is exactly why does ChatGPT make up references is worth reading before a single generated citation goes anywhere near a reference list.

The reasoning behind all of this is more consistent than the rules themselves. A chatbot cannot stand behind a claim it generated, check it against a source, or take responsibility if the claim turns out to be wrong, which is exactly why research publishers require a human author to vouch for every claim regardless of what helped produce the sentence. Whether Turnitin's AI writing report would flag a specific paragraph is a separate, mechanical question from whether the underlying content was ever the student's own to submit.

The Grey Middle, and How to Handle It

Some uses sit between the two clear categories above, and this is exactly where a course-specific policy matters most. Translating a difficult passage, or checking whether a sentence reads naturally in a second language, sits differently for a multilingual student than for one writing only in a first language, and a policy that does not mention this at all has usually just not been written with that student in front of the person drafting it. Asking a tool to explain a concept before writing about it in original words is closer to using a textbook than to generating a submission, but not every marker would agree, which is precisely why this is the category worth asking about directly rather than guessing.

Brainstorming a list of possible angles, or asking for a rough outline before writing every sentence from scratch, falls into the same grey space. Some instructors treat it as no different than talking an idea through with a classmate. Others want every stage of a submission to be visibly the student's own. The only reliable way to know which applies to a specific assignment is the same one this guide opened with: ask, or read the brief closely enough to be sure.

Keep a Record Without Turning It Into a Second Assignment

A simple habit covers most of this without becoming a project of its own. Note the date, what was asked of the tool, and what was actually kept versus rewritten, the same way a source gets a citation rather than a vague memory of having read something once. Save the chat if a course allows it, or a screenshot if it does not. Where a course requires a formal AI-use disclosure statement, that format is set by the course, not by this guide, and following its exact wording matters more than writing a longer or more thorough one of independent design.

The practical questions underneath the principle have their own pages: whether a professor can tell you used ChatGPT, and the blunter version, will I get caught using ChatGPT.

None of this replaces reading the one document that actually governs a specific assignment: the module handbook or the brief itself, not a general impression of what other students seem to be getting away with. TextPulse's AI humanizer for students exists for the part of this that is genuinely about writing, not about permission. It reports an estimated Human Score built from a draft's own statistical shape, useful for checking authentic writing before submission, on exactly the categories of use a policy already allows.