5 prompts for self-study that actually work.
Most study-prompt lists are useless because they ask the model to summarise instead of teach. These five force active engagement and have a checkable end state.
Most "prompts to study X" lists are useless. They are vague ("explain quantum mechanics like I'm 5"), confuse complexity for depth, and produce surface answers you forget by next week. This page is the opposite: prompts that work because they force the model to do the work of a tutor, not the work of a Wikipedia summariser.
The principle
A good study prompt does three things:
- Tells the model your current level honestly.
- Asks for active engagement: questions, examples, problems, not just exposition.
- Sets a checkable end state so you know when you are done.
Every prompt below follows that shape.
1. The Socratic walkthrough
I want to understand [topic]. My current level: [describe what you already know in one sentence]. Ask me one question at a time. After my answer, correct anything wrong, then ask the next question. Build up from the basics. Tell me when you think I understand it well enough to teach it back to you.
When to use it. A new concept where you have surface knowledge but cannot yet explain it. Especially good for maths, security concepts, networking, anything systematic.
Pitfall. If you skip questions or guess your way through, the model has no signal to correct you. Be honest.
2. The flashcard generator
From the following notes, produce 20 spaced-repetition flashcards. Format: Q on one line, A on the next. Atomic answers only – one fact per card. No essay answers. Skip anything I already obviously know. Notes: [paste your study notes here]
When to use it. Right after a lecture, a chapter, or a working session. The model will not produce magic recall, but it will turn 40 lines of messy notes into 20 cards in 30 seconds.
Pitfall. Verify the cards before you import them anywhere. LLMs hallucinate facts, especially numbers and dates.
3. The exam-style problem generator
Generate 5 problems on [topic] at the level of [name of textbook, exam, or course]. Mix easy, medium, and hard. Do not give solutions yet – wait for me to attempt each one. After I send an attempt, mark it, point out the error if any, and only then give the full solution.
When to use it. Revising before an exam, or when you need to know whether you can do the thing, not just whether you can recognise it.
Pitfall. If the model invents problems that are too easy or too unrealistic, change the reference. "At the level of MIT 6.006 problem set 3" is a sharper anchor than "at university level".
4. The "explain my mistake" prompt
I attempted this problem and got it wrong. My working: [paste your working]. The correct answer is [paste it]. Walk me through exactly where my reasoning broke. Do not solve the problem from scratch – I want to see my error in context.
When to use it. When a textbook gives you a one-line answer and you cannot work out where you went wrong. The model is good at this because it does not have to be creative: it just has to compare two trajectories.
Pitfall. If your working is illegible (handwritten and badly typed), the explanation will drift.
5. The "what should I learn next" prompt
My goal is to [goal in one sentence]. I currently know: [bullet list of what you know]. Identify the smallest next thing I should learn to move toward the goal. Justify your pick in one paragraph. Then suggest one free resource for it.
When to use it. Anytime you are at a fork and tempted to start three things at once. The constraint to one next thing is doing the work.
Pitfall. Verify the resource exists before you commit. The model may invent a paper title or a course URL.
Privacy and data sensitivity
If your study notes contain personal data (your employer's internal docs, a classmate's name, anything that identifies a person), strip that before pasting. Most chatbots log conversations and use them for evaluation or training unless you opt out. Treat paste boxes like public Slack channels until proven otherwise.
Where to go next
These five cover 80% of self-study. The interesting frontier is not "better prompts": it is habits. The Socratic walkthrough does nothing if you only run it once. Set a recurring time. Treat the chatbot as a study partner with limited memory, not an oracle.
Daniel A. and Óscar S. run Breachfolio, a small independent site about security and AI. This article was drafted with AI assistance and reviewed by a person before it went live. We write from documentation, vendor sources and published research rather than from original lab benchmarks, and we link a source in the sentence that relies on it. How we work · About us