An Instagram carousel crossed my feed this week claiming Claude has a mode called “Infinite Private Tutor” that teaches you any skill from zero in four hours. It doesn’t. There is no such mode, and four hours gets you to dangerous-at-a-dinner-party, not to skilled.
The carousel is eigenlijk better than its headline. Underneath sit six decent prompts from @quentin_aimarketing, and they share one idea worth stealing: the useful version of AI as a tutor isn’t the one that explains things. It’s the one that refuses to.
Why explanations don’t stick
Ask any chatbot to explain something, and you get a tidy, confident answer. You nod, you feel smarter, you close the tab. Two days later it’s gone. Reading an explanation feels like learning in the same way watching a cooking programme feels like cooking.
What the six prompts have in common is that they flip the roles. You do the work. The AI sets the task, waits, and checks. That is how a good teacher operates, and it’s the opposite of how most people use ChatGPT or Claude.
Start with the result, not the subject
The first two prompts are about planning, and the first one fixes the most common mistake: asking to “learn SEO” or “get better at Excel”. That’s a subject, not a goal, and the AI will happily give you a syllabus for a subject. Instead, tell it the specific result you want and the deadline. “I want three of my category pages on the first page of Google by December” gives it something to plan backwards from. Add what you already know, then ask for a seven-day path: one task a day under 45 minutes, a test so you know whether you got it right, and one thing to skip.
That skip is the most valuable line in the whole prompt. Anyone can tell you what to learn. Knowing what to ignore is where an expert earns their fee.
The second prompt adds pressure. Tell the AI it has four hours with you and will never see you again, and its only job is to make you functional. Ask what to learn first, what to ignore completely, and which single exercise would put you ahead of people who have been at this for months. Then the important bit: have it teach step one and wait for your reply. Without that instruction, you get a wall of text, which is a textbook, not a tutor.
Make it ask, not tell.
The middle two prompts are where the real learning happens.
For anything that confuses you, paste it in and ask for the one core idea that makes the rest fall into place. Have that idea explained with an everyday analogy and no technical terms. Then ask for three questions that only someone who truly understood it could answer, one at a time, and tell the AI not to move on until you pass all three. It’s a small change, but it turns reading into proving.
The mistake simulator goes further. Don’t ask for an explanation. Ask to be dropped into a realistic situation where you’ll have to use the concept and will probably get it wrong. When you do get it wrong, the AI shouldn’t hand you the answer; it should ask one question that helps you find where your reasoning broke. It only answers two attempts, then repeats with a new situation until you get it right without hesitating. Flight simulators work the same way. Nobody learns to land a plane by reading about landing.
Then try to break what you think you know.
The last two prompts are for after you’ve studied something, which is exactly when you’re most likely to overestimate yourself.
The first is the Feynman method with a referee. Tell the AI you’ll explain what you learned as if to a ten-year-old, and ask it to stop you whenever you use jargon you can’t define, skip a step, or oversimplify until it becomes wrong. At the end, it tells you what those slips reveal about the shaky parts of your understanding.
The second is blunter: tell the AI you think you’ve mastered the skill and ask it to prove you wrong. Five questions that look simple but expose anyone who never went deep, one at a time, with feedback after each on what’s missing from your foundations. Tell it not to go easy on you. Use this one when you’re feeling pleased with yourself. It is very good at fixing that.
What to watch out for
An AI tutor has one weakness a human tutor doesn’t: it will sometimes confidently mark a wrong answer as right, especially in niche subjects. These prompts work best where you can check the result against something real, such as code that runs or doesn’t, a spreadsheet that adds up or doesn’t, or a source you trust.
Models also drift back towards being nice. If the feedback starts sounding like a participation certificate, tell it again to be direct. And “wait for my reply” only works if you actually reply, rather than typing “continue” because you want to see where it’s going.
The point
There is no tutor mode. You create a mode yourself: be specific about the result, make the AI wait, and let it catch you when you’re wrong. Four hours won’t make you an expert in anything. But four hours of being corrected will get you further than forty hours of being explained to.
