The honest version of what AI is good for at university
Two things are true at once. AI is the most useful study tool that has ever existed, and using it the obvious way will make you worse at your subject. The difference is entirely in whether the model does the thinking or interrogates yours.
Asking for an essay produces a passable essay and zero learning. Asking a model to quiz you on a chapter until you can explain it without notes produces no artifact at all and a genuine grasp of the material. Both take the same amount of time. Only one of them helps in the exam, where you will not have it.
Everything below is built around that distinction. The workflows that hold up are the ones where you generate first and the model reacts, rather than the reverse.
| What you need | Model | Why |
|---|---|---|
| Summarizing readings, slides, long PDFs | Gemini | 1M token context, so a whole textbook chapter or lecture set fits at once |
| Essay structure, counterarguments, outlines | GPT | Strong at explicit structure and at holding an argument to a brief |
| Prose, feedback on your writing, tone | Claude | Best writing quality and the most useful critique of a draft |
| Debugging code assignments | GPT or Claude | Both explain errors well; compare them when one is unconvincing |
| Active recall and practice questions | Any of them | Quality depends on the prompt far more than the model |
Turning readings into something you will actually revise from
A summary you did not write is a summary you will not remember. The fix is to make the model produce revision material that requires you to do the retrieval.
Prompt: the study guide
I have uploaded [the reading]. Produce: (1) the five to eight core concepts with a one sentence definition each, in the source's own terminology; (2) how the concepts connect to each other; (3) the three points a first-time reader is most likely to misunderstand; (4) ten questions that test understanding rather than recall, without answers. Do not include anything not present in the document.
Withholding the answers is the whole trick. Answer the ten questions yourself first, then paste your answers back with: Mark these against the source. For each, say what is correct, what is missing, and what is wrong. Do not be generous.
Prompt: the Feynman check
I am going to explain [concept] in my own words. Identify every place where my explanation is vague, circular, or wrong, and ask me one follow-up question that would expose the gap. Do not explain it for me. My explanation: [type it out from memory].
That prompt is worth more than any summary on this page. Explaining from memory and getting a targeted objection back is close to what a good tutorial with a supervisor does, and you can run it at midnight the day before.
Prompt: exam-shaped practice
Generate [n] exam questions on [topic] at the level of [course and year]. Match the style of these past questions: [paste two or three]. Do not provide answers. After I answer, mark against the criteria that would apply, state the grade band, and give the single change that would move it up one band.
Essays: where the line actually is
Submitting model-written prose as your own is plagiarism at essentially every institution, and it is also the least useful thing you can do with the tool. Everything short of that is a spectrum, and most departments have published where they draw it. Read your own department's policy, because they differ more than students assume.
What tends to be both permitted and genuinely useful:
- Interrogating your thesis before you commit.
Here is my argument. What is the strongest objection to it, what evidence would I need to defeat that objection, and is there a version of this claim that is more defensible? - Structure from your own material. Give it your notes and your argument, and ask for two or three possible structures with the trade-off each makes.
- Feedback on your draft.
Mark this against these criteria: [paste rubric]. For each criterion, quote the strongest and weakest passage and say specifically why. Do not rewrite anything. - Finding the gap.
Which claims in this draft are asserted without support? Quote each one. - Explaining a source you are stuck on. Upload it and ask for the argument structure, then check the explanation against the text yourself.
What to avoid regardless of policy: generating paragraphs and editing them into your voice (that is still not your argument), and asking for sources without opening every one. Models fabricate citations that look completely real, with plausible authors, journals, and page numbers. A fabricated reference in a bibliography is caught easily and treated seriously.
Prompt: the rubric pass
Here is my essay and the marking criteria. For each criterion: the grade band this draft currently sits in, the exact passage that is holding it there, and the single most efficient change to move it up. Be blunt. Do not rewrite my sentences. Essay: [paste]. Criteria: [paste].
Coding assignments, without breaking the point of the assignment
The temptation is to paste the problem and copy the answer. That works until the exam or the technical interview, at which point the gap is unpleasant. The alternative is barely slower and actually teaches.
Prompt: explain the error, do not fix it
Here is my code and the error. Explain what the error means, why it is happening in this specific code, and what category of mistake it is. Do not give me the corrected code. Ask me what I think the fix is.
Prompt: review after you have it working
This works. Review it as a demanding TA would: correctness edge cases, complexity, naming, and anything that would lose marks on style for [language and course]. Rank the issues by how much they matter. Do not rewrite it.
Prompt: the concept behind the bug
I keep making this mistake: [describe]. What is the underlying concept I have not understood? Explain it, then give me three small exercises that would test whether I have it, without solutions.
When one model's explanation does not land, switch and ask the other. Claude and GPT explain the same concept with genuinely different framings, and the second framing is often the one that clicks. That single habit is the most concrete argument for having more than one model available.
What it costs, and the honest comparison
ChatGPT Plus is about $20 a month for the GPT family alone. Claude Pro is another $20. Gemini through Google One AI Premium is another $20 again. Paying for all three as a student is not reasonable, and picking one means losing the specific thing the other two are better at, which for coursework is exactly the point.
Whizi's Starter plan is $15.99 a month, or $10.99 a month billed yearly, below any single one of those, and covers GPT, Gemini Flash, and Whizi AI, with Pro adding Claude Sonnet, image generation, and higher limits. The savings calculator will do the arithmetic against whatever you are currently paying.
The free tiers of each product are a legitimate option too, and worth saying plainly: if your usage is light, free accounts across two providers will cover a lot. What you hit is rate limits during exactly the weeks you need it most, which is the week before a deadline.
Six habits that separate students who get better from students who get faster
- Always attempt first, then ask for critique. Generating before the model does is the difference between practice and consumption.
- Ask for questions rather than answers, and answer them before revealing anything.
- Upload the source instead of asking what the model remembers about it. Memory-based answers about specific papers, cases, and texts are unreliable.
- Never submit a citation you have not opened.
- Switch models when an explanation does not land, rather than re-reading the same one.
- Write down what you actually did. Many courses now ask for an AI use statement, and reconstructing it later is worse than noting it as you go.
- Pick your hardest module and upload the core reading for it this week
- Ask for a study guide with questions and no answers, then answer them from memory
- Run the Feynman check on the concept you are least confident about
- Get rubric-based feedback on your next essay draft instead of asking for a rewrite
- Ask for explanations of code errors rather than corrected code
- Check your department's AI policy before the deadline, not after
- Open every source before it enters a bibliography
Frequently asked questions
Is there a student discount?
Whizi’s Starter plan is already priced below a single ChatGPT Plus seat while covering GPT, Gemini Flash, and Whizi AI. Claude starts on Pro. Check the pricing page for current promotions and any student offer running at the moment, and the calculator if you want the comparison against what you already pay.
Can I use Whizi without internet for exam prep?
You need a connection to chat with the models, since the models run on the providers’ infrastructure rather than on your device. The practical workaround is to generate what you need in advance and export it: study guides, practice question sets, and worked explanations all copy out as text you can revise from offline.
Will professors know I used AI?
Detection tools exist and they are unreliable in both directions, producing false accusations and missing real cases, which is why most institutions have moved toward disclosure requirements rather than detection. That is the practical answer, but not the important one. The important one is that submitting model-written work is plagiarism at nearly every institution regardless of whether it is detected, and it leaves you without the understanding the assessment was measuring. Whizi is not a detection-evasion tool and there is nothing on this page that is.
Which model is best for essays?
Claude, for feedback on your writing and for the clearest critique of an argument. GPT is better when you want structure, counterarguments, or an outline held tightly to a brief. Use Gemini for the reading stage, since it can hold an entire set of course materials in one conversation. The workflow most students settle on is Gemini to read, GPT to structure, Claude to critique.
Can it read my lecture slides and textbook chapters?
Yes. Upload PDFs, slide decks, and documents directly, and Gemini in particular can handle a very large amount at once, so a full set of lecture slides or several chapters can sit in a single conversation. Always upload the source rather than asking what the model already knows about a text, because answers drawn from memory are much less reliable than answers drawn from the document in front of it.