An AI Tutor That Makes You Think Instead of Lecturing

Some people use ChatGPT as a private tutor to learn new skills: instead of having it lecture, they have it give hard problems, wait for you to make mistakes, probe your reasoning, and finally become your student. This approach has spread widely online and is claimed to get you started on a skill in a few hours. The timing claim is a bit exaggerated, but the direction is right: AI is most useful as a teacher not when it explains things clearly, but when it makes you think.

How this private tutor divides the work

The original is eight prompts. We have grouped them by learning order into five roles, each of which appears only at one stage:

Role	When to use it	What it makes you do
Planner	At the start, when the goal is still vague	Turns "learn a skill" into a result you can test within a deadline, and lays out small tasks for each day
Breaker	When you can't understand a piece of material	First finds the one sentence that, once understood, makes everything else clear, explains it with everyday analogies, then quizzes you to confirm
Quizmaster	After learning a concept	Gives you a realistic scenario where it's easy to go wrong; if you get it wrong, it gives no answer and keeps asking questions until you find the error yourself
Auditor	When you think you've got it	Uses seemingly simple questions to find your gaps; even when you answer correctly, it asks how you arrived at the answer
Student	The final stage	You teach it; it pretends not to understand and zeroes in on the points you can't explain clearly

ChatGPT, Claude, and Gemini can all be used this way without relying on any hidden feature.

Why it works: three reasons

• Make mistakes first, then get the answer: Thinking something through, getting it wrong, and then seeing the correct method sticks far better than simply listening. That is why the quizmaster is told "don't give the answer when I'm wrong."
• Only count correctness when you can explain why: An answer may be right by guessing or memorizing. The auditor specifically checks your reasoning, separating "can do it" from "truly understand."
• Teaching someone else is the real test: Explaining to a "student" who asks follow-up questions reveals what you haven't understood yet, more easily than reviewing on your own.

Learn along with it: a five-step SOP

Give this table to your AI and have it guide you step by step, checking yourself against the "completion standard" after each step.

Common pitfalls

• "Learn any skill in a few hours" is an exaggeration: This method can make getting started much faster and learning more solid, but skills that require lots of practice (instruments, sports, real-world programming) still take time.
• Tracking progress across conversations depends on the memory feature: The original includes a line asking the AI to "remember my previous answers," which only works if memory is turned on or you stay in the same conversation. A more reliable approach is to have it write a progress summary at the end of each day and paste it back at the start of the next.
• The AI's quiz questions and answers can also be wrong: Specialized content (medicine, law, finance) should be checked against authoritative sources, and its judgments should not be treated as the official answer.
• Don't use all five roles from the start: First run the "planner" and "quizmaster" for two or three days, and add other roles only when you get stuck.
• "No answers" can be frustrating: This is intentional. If you're stuck for more than 10 minutes, you can ask it for a hint rather than the answer outright.