Using AI to study without ending up in an integrity case
The rules are set per course, not per campus. And the failure mode isn't getting caught — it's arriving at the exam having watched someone else do the work.
·5 min read
Two separate problems get mashed together in most advice about this, and they have different answers.
One is am I allowed to do this. The other is does this actually help me learn. You can be entirely within the rules and still walk into a midterm having learned nothing, which is the outcome nobody warns you about because everyone's busy talking about cheating.
The rules are per course. There is no default.
There's no national standard and, at most schools, no campus-wide rule. Policy is set in the syllabus, by the instructor, course by course.
In one semester you can easily have four courses with four different rules — a writing course that prohibits it entirely and runs handwritten in-class essays, a psych course that allows it for outlining but not drafting, a stats course that allows it with disclosure, and a CS course where using it well is part of the curriculum.
The variation isn't random, either. A survey of syllabi across many disciplines found no computer science instructor imposing a total ban, while writing-intensive humanities courses ban it far more often. That makes sense — the thing being assessed is different.
Three practical rules follow.
Read the AI paragraph in every syllabus, every semester. Policies are still changing fast. Last year's rule for the same course may not be this year's.
If the syllabus doesn't say, that isn't permission. It's a question you haven't asked. One line of email: "Is AI assistance permitted on the problem sets, and if so what should I disclose?" Nobody has ever gotten in trouble for asking.
When a course requires disclosure, disclose properly. A sentence at the end saying what you used it for. Vague disclosure reads worse than none.
The part that isn't about rules
Here's what I think is the more common failure, and it's completely legal.
You're stuck on a problem. You ask, you get a clear worked solution, you read it, it makes sense, you move on. Twenty minutes saved.
What just happened is that you watched someone solve a problem. The clarity you feel is recognition — the same thing that makes rereading feel productive and makes practice-exam solutions look obvious before you've attempted them. You did not do the retrieval, you did not sit in the discomfort of not knowing, and neither of those is available to you on exam day.
The exam is the only moment where nothing explains anything to you, and it's the moment that counts.
This is worth being blunt about: the risk isn't mainly getting caught. It's arriving at the exam having outsourced the exact skill being tested.
Uses that work
The distinction I use is whether the tool is making me produce something or producing it for me.
Make it quiz you. Paste your lecture notes and ask for ten questions, no answers. Answer them cold. Then ask for the answers. This is retrieval practice with the question-writing outsourced, which is the boring part.
Explain your understanding and have it find the holes. Write out your explanation of a concept, then ask what's wrong or missing. You produced first; the feedback comes second. Same shape as explaining it to a roommate, minus the roommate.
Ask for a different explanation of something you already read. When the textbook's framing isn't landing, a different angle genuinely helps. Verify against the textbook afterward, because a plausible wrong explanation is worse than a confusing right one.
Ask it to generate practice problems of a type you're weak on. Then do them without help. Check the answers against your own work, skeptically.
Debug your reasoning, not your answer. "Here's my approach, here's where I got stuck, what am I missing" preserves the part where you were doing the thinking.
The pattern in all of these: you produce, then it responds. Reverse that order and the value collapses.
Uses that hurt
Getting solutions before attempting. The single most damaging habit. The struggle is not an obstacle to learning, it's the mechanism.
Having it write anything you submit. Rules aside, if the assignment was writing and you didn't write, the assignment did nothing for you.
Trusting it on facts, dates, formulas, or citations. More on this next.
Summarizing a reading you then discuss in class. You'll be fine for ten minutes and lost the moment someone asks about something the summary flattened.
The fabricated citation problem, specifically
If you ask for sources, a meaningful share of what comes back may not exist. Plausible authors, plausible journals, plausible years, no such paper.
This is measurable and it's getting worse in the published literature — analyses tracking fabricated references in academic papers have found the rate climbing sharply over the last few years. In law it has produced repeated court sanctions against attorneys who filed briefs citing cases that were never decided, with penalties running to thousands of dollars.
For a student paper this is treated more severely than a formatting error, and reasonably so. A missing citation says you were careless. A citation to a nonexistent source says you never read your sources — and a grader can check that in about ninety seconds.
Never cite anything you haven't opened. Not the abstract. The source. If your library can't find it, assume it isn't real.
Detection, and why it's not the point
Instructors have AI-detection tools and they're unreliable in both directions — false positives on unusual-but-human writing, false negatives on lightly edited output. The current guidance most schools give faculty is to treat a flag as a signal to look closer, not as proof.
That's genuinely bad news for you if you're falsely flagged, and the defense is process: keep drafts, keep version history, keep your notes. Being able to show how a paper came to exist is worth more than any argument about the detector.
It's also why blue books came back. Sales have jumped across large campuses over the past two years as instructors moved graded writing into the room, where the question of detection doesn't arise. If your syllabus mentions handwritten in-class essays, that's the reason — and it means practicing writing by hand, under time, is now actually part of preparing.
The one-line test
Before you paste a question in, ask: am I about to skip the part I'll be graded on?
Skipping the tedious part is fine. Skipping the part that constitutes the skill is how you end up in an exam room realizing you've spent a semester watching.
Keep reading
Pomodoro didn't work for me either
Twenty-five minutes is great for work made of small pieces and actively bad for work you have to hold in your head. The timer isn't the problem.
Where did the week 3 slides go
It's finals week and the handout you need is in a group chat that has scrolled past it. The fix is one rule, applied on the day things arrive.
Access codes: how not to pay for the same book twice
You bought the used copy and saved sixty dollars. Then the graded homework turned out to live on the publisher's site, and the code costs full price.
Nobody goes to office hours. That's the whole opportunity.
Surveys keep finding that most students never use them. Meanwhile the professor is sitting in an empty room for two hours a week, on purpose, for you.