The best ChatGPT prompts for university students in 2026

Before uploading, check permissions and remove personal or restricted information. For assessed work, check whether feedback, rewriting or other AI assistance is allowed. A prompt cannot override those rules.

• Ask the tool to say when the supplied material is insufficient.
• Open cited passages and check calculations or generated answer keys.
• If feedback conflicts with your course, use the official source or ask your tutor.

These prompts can also be adapted for Claude or Gemini. Interface features and file access depend on the tool and account.

Most students ask AI to explain the whole topic. Don't. First find the part you actually don't understand.

I'm going to explain [TOPIC] in my own words. Do not teach it to me yet. Read my explanation and identify the first point where my understanding becomes incorrect, incomplete or vague. Don't fix everything at once. Ask me one question that would help you determine what I actually misunderstand.

It stops the AI giving you another perfectly written explanation of things you already know. The tutoring begins at the edge of your understanding.

Sometimes the reason slide 37 makes no sense is slide 12.

Before you explain [TOPIC], identify the most important prerequisite concept I need to understand first. Ask me two short questions to test whether I understand that prerequisite. If I do, continue to the main topic. If I don't, teach me only the missing prerequisite first and then test me again.

This is particularly good for maths, economics, statistics, physics, chemistry, accounting and programming. Difficult topics usually sit on top of easier ones. AI is much more useful when it helps you find the missing brick rather than explaining the entire wall.

OpenAI’s study workflow uses course material to guide practice. You can also request a conversational tutoring approach explicitly, though the model may not follow every instruction:

Tutor me on [TOPIC]. Do not give me long explanations unless I need them. Ask one question at a time. Use my answer to decide what to ask next. If I'm wrong, don't immediately tell me the answer — give me the smallest useful hint. Keep going until I can explain the topic correctly without your help.

Check the questions and feedback against your course. A confident explanation from ChatGPT can still be wrong. The aim is for you to explain the topic accurately after the help is removed.

One of the easiest ways AI tutoring goes wrong at university is by giving you a perfectly valid explanation that doesn't match your module. Different lecturers use different notation, methods, definitions, assumptions, examples and conventions. Upload the lecture and use:

Teach me [TOPIC] using the attached lecture material as the primary source. Preserve the lecturer's notation, terminology and method. Start with the intuition, then connect that intuition to what appears on the slide. If there is a common alternative method that is not taught in these materials, don't silently switch to it — mention that it exists, but keep the explanation aligned with my course. Finish by asking me one question that tests whether I understood the actual slide rather than your explanation.

Instead of “make me a quiz”, use:

Using the attached material, test me on the most important ideas one question at a time. Do not show the answer until I respond. Prioritise questions that require me to retrieve or apply information rather than recognise it from multiple-choice options. After each answer: tell me what I got right, tell me exactly what was missing or incorrect, ask a harder follow-up if I understood it, and revisit the idea later if I did not.

This turns a static document into a feedback loop — much more useful than reading the same highlighted paragraph for the sixth time.

This one is excellent before an exam.

I think I understand [TOPIC]. Try to prove that I don't. Start with a normal university-level question and make each question harder based on my answers. Include edge cases, applications and questions that distinguish genuine understanding from memorising definitions. Stop when you find the boundary of my understanding. Then tell me exactly what I should revise.

There is something psychologically useful about changing the goal. You're no longer trying to convince yourself that you know it. You're actively looking for the thing that could expose you in the exam. Better to find it now.

“Improve my essay” is one of the worst prompts if your goal is actually to become a better writer. It lets the model solve every problem simultaneously.

Read this as a strict university marker. Do not rewrite any sentences. Identify the three issues most limiting the quality of the argument. Rank them by how much fixing each one would improve the work. For each issue: show me where it appears, explain why it weakens the argument, and ask me a question that forces me to improve it myself. Do not propose replacement wording unless I explicitly ask.

Now the feedback belongs to you. And so does the revised argument. Whether this is permitted for a given piece of assessed work is a separate question — check your university's AI rules.

AI assistants are often extremely good at helping you develop an idea you've already chosen. That can become a problem.

Here is my argument: [ARGUMENT]. Assume an intelligent reader strongly disagrees with me. Give me the strongest reasonable counterargument they could make. Do not invent fringe objections just to disagree. Identify the evidence or assumption in my argument that this counterargument attacks. Then ask me to respond before you evaluate whether my response works.

Useful for essays, dissertations, presentations, business cases, research proposals and seminar preparation. Agreement feels good. Pressure-testing is usually more valuable.

Students often know two definitions individually but still confuse them in an exam.

Help me distinguish [CONCEPT A] from [CONCEPT B]. Don't just give me two definitions. Explain what they have in common, the precise difference that matters, a case where both might initially seem applicable, and how I would decide which one applies. Then give me three examples one at a time and make me classify them.

This works brilliantly for concepts that live too close together in your head: correlation and causation, fiscal and monetary policy, Type I and Type II error, negligence and strict liability, sympathetic and parasympathetic. Knowing definitions is not always the same thing as being able to discriminate.

If you paste a problem into AI and read the solution, you've learned something. Possibly. If you want a better chance:

I want to solve this problem myself: [PROBLEM]. Do not solve it for me. Ask me what I think the first step should be. If I'm wrong, diagnose why and give me the smallest hint needed to continue. Keep the solution aligned with the methods and notation in the attached course material. Only show a complete worked solution after I have either solved it or explicitly given up. At the end, give me a similar problem with different numbers or assumptions and no help.

The last part is the test. Understanding someone else's solution can create a very convincing illusion that you would have produced it yourself. A fresh problem reveals whether you actually could.

One of the best prompts when you're working in an unfamiliar area.

Before we continue, what important questions, assumptions or pieces of evidence are missing from my analysis? Rank them by how much they could change my conclusion. Don't list generic improvements. Focus on things I appear not to have considered at all. Then tell me which three I should investigate next.

This works for far more than studying — research, dissertations, business plans, case studies, presentations, interview preparation, data analysis. The questions you already know to ask are rarely the dangerous ones.

This may be the most useful prompt on the list.

We've been studying [TOPIC]. I now want to test whether I actually learned it. Ignore how well I followed your explanations. Give me a short notes-closed test covering the most important things I should now be able to do. Test understanding and application, not just vocabulary. Ask one question at a time and do not provide hints unless I request them. At the end, separate the results into things I can now do independently, things I partially understand, and things I still need to learn.

This solves one of the biggest problems with AI-assisted studying. AI can make everything feel clear. The explanation is always available. The next step is always suggested. The missing equation is always filled in. Then you close the laptop — and discover how much of that clarity belonged to the model.

Suppose you tell the tool: “Doubling the sample size halves the standard error.” Under the formula SE = σ / √n with σ held fixed, that statement is wrong: doubling n multiplies SE by 1 / √2. You need four times the sample size to halve it.

Helpful feedback identifies the square root and gives you a small follow-up, such as calculating the change from n = 25 to n = 100. Unhelpful feedback praises the statement or explains sampling without addressing your mistake. This is an illustrative answer check, not a transcript of a model test.

Before using a prompt on an unfamiliar topic, try it on a claim you can verify. If it misses an error you already know about, do not rely on its judgment uncritically.

You might have noticed something. Very few of these prompts ask the AI to produce more content. They ask it to diagnose, question, challenge, test, compare, give feedback and find gaps.

That is deliberate. The easiest way to use ChatGPT at university is “do this.” The more interesting way is “make me better at doing this.” Those sound similar. Across a three-year degree, they compound in completely different directions — which is the same argument as AI study tools versus traditional note-taking.

If you only save one prompt, save this:

Don't solve this for me yet. First work out what I already understand, what I don't understand, and what the smallest next thing is that you could help me learn. Then teach only that.

That is probably closer to what most students actually want from an AI tutor. Not more information. The right help at the point where they got stuck.