Using ChatGPT to study: what it is good at, and what it quietly ruins
An AI assistant will summarise a chapter in seconds, and reading that summary feels like learning. It is not, and the gap between those two things is the single most important thing to understand before using one for a degree.
Anything that produces the answer for you has removed the part that does the learning.
The one rule that decides everything
Retrieval is what builds knowledge — producing something from your own memory before checking it. Anything that hands you the produced thing skips exactly that step.
So the useful split is not between allowed and forbidden. It is between using the model to make you produce more, and using it to produce instead of you. Summaries, essay drafts and worked answers all sit on the wrong side of that line, however good they are.
What it is genuinely good at
Being examined by. "Ask me ten questions on this chapter, one at a time, and do not tell me the answer until I have tried" turns a passive text into retrieval practice, which is the highest-value study activity there is.
Explaining something at a level you choose. "Explain this as you would to someone who understands limits but not continuity" produces a bridge from what you know, which is the shape understanding actually takes.
Being explained to. Ask it to check your explanation of a concept and name what is missing. This is the Feynman technique with a listener, and it exposes gaps in a way that reading never does.
Generating practice problems in a form your course uses, which is the tedious part of self-testing and the reason people stop doing it.
Where it quietly ruins things
Summaries. A summary you did not write teaches you almost nothing and produces strong familiarity, which is the exact false signal that makes re-reading feel productive. It is the most popular use and the least useful one.
Anything with a right answer you were meant to work out. Getting the solution and following it produces recognition of the method, and recognition collapses under exam conditions.
And it is confidently wrong at a rate that matters. In any subject where you cannot yet tell a correct explanation from a plausible one — which is the definition of a subject you are learning — you are not in a position to catch it.
Four prompts worth keeping
"Quiz me on this material one question at a time. Do not give me the answer until I answer. After each one, tell me only whether I was right and what I missed."
"Here is my explanation of X. What is wrong or missing in it? Do not rewrite it for me."
"Give me five problems on X at the level of this example, without solutions. Hold the solutions until I ask."
"I am stuck on this problem. Give me the smallest hint that would unstick me, not the next step."
The pattern in all four is the same: the model is instructed to withhold. Left to itself it will help too much, because helping is what it is built to do.
The academic honesty part
Using an assistant to be quizzed, to have something explained, or to check your own explanation is study. Submitting text you did not write is misconduct at most institutions regardless of how it was produced.
The line is your institution's, not this page's, and it is worth reading rather than assuming — the rules differ significantly between universities and sometimes between departments.
FAQ
Is using ChatGPT to study cheating?
Being quizzed by it, having a concept explained, or asking it to critique your own explanation is studying. Submitting text you did not write is misconduct at most institutions however it was produced. The rules differ between universities and departments, so read yours rather than assuming.
What is the best way to use ChatGPT for studying?
Instruct it to withhold. Ask to be quizzed one question at a time with the answer held back, ask for the smallest hint rather than the next step, and ask it to find the gaps in your own explanation instead of writing one for you.
Should I use AI to summarise my lecture notes?
It is the most popular use and one of the least useful. A summary you did not write produces strong familiarity and very little recall — the same false signal that makes re-reading feel productive. Writing the summary yourself is where the value was.
Can I trust what it tells me?
Not unverified. It is confidently wrong often enough to matter, and in a subject you are still learning you are by definition not able to tell a correct explanation from a plausible one. Check anything that will end up in an answer against the course material.
An assistant can quiz you; it cannot tell you what is due on Thursday. majored keeps courses, deadlines and exams in one place on Mac, iPad and iPhone, with a focus timer for the sessions themselves.