Methods · Dialogue, argument & inquiry
C1
Socratic questioning
Turn an AI from something that explains things to you into something that makes you explain things to it, and find out, in about ten minutes, which parts of a topic you only think you understand.
When to use it
- You can recite the definition, but you freeze when someone asks why.
- A seminar or tutorial is coming and you need your reasoning under pressure.
- You feel fluent after reading, then blank when someone asks you to explain without notes.
What to ask AI
Type something like: “You are interviewing me about [topic]. Ask one question at a time and wait for my answer.”
Tell it not to explain, correct, or add information until you ask.
Keep the hard part on your side: you answer before you look anything up.
How it helps
You can read a chapter three times and feel it getting easier each time.
That feeling is fluency, and it is not understanding.
It comes from recognition: the words look familiar now.
It disappears the moment someone asks you to explain the idea without the page in front of you.
Socratic questioning attacks that gap.
Instead of receiving an explanation, you produce one, then defend it against follow-up questions that probe the joints of your reasoning.
The load-bearing idea is explaining in your own words: generating an answer beats re-reading the same text.
That result is solid for human learning; whether an AI interviewer beats a worksheet or a tutor is a separate, still-open question.
How to run it
- Name a target small enough to be wrong about. “Photosynthesis” is too big; “why plants need light, not just CO₂ and water” is workable.
- Write your current understanding in three or four sentences before any questioning. Without that, you will reshape your view to match the questions.
- Let the AI ask one question at a time. Five at once means you answer the easy ones and skip the rest.
- Answer before you look anything up. A wrong answer you generated is worth more than a right one you read.
- When you are vague, make it ask what you mean. When you assert, make it ask how you know.
- Keep going until you hit a question you cannot answer. That breaking point is the destination.
- Close the gap in your textbook or notes, then re-run the same interview and see whether the breaking point moved.
Example exchange
You said a limit is “what the function gets close to.” Consider a function that equals 1 everywhere except at x = 2, where it equals 5. What is the limit as x approaches 2?
Five? No, wait, one, because everywhere near 2 it is still 1.
So does the function’s value at the point itself matter to the limit?
No. The limit is about the neighbourhood, not the point.
Copyable prompt
You are interviewing me about [TOPIC]. Ask ONE question at a time and wait for
my answer. Do not explain, correct, or add information. When my answer is vague,
ask what I mean by the vague word. When I assert something, ask how I know it.
Start with a question a beginner could answer and go deeper based on what my
answers reveal I don't know. Begin.The Tell
If you end the session having learned facts the AI supplied rather than gaps you discovered, it was not Socratic questioning. It was a lecture with question marks.
The tell is simple: you never stalled, or you stalled and then asked the AI to fill the blank.
Tighten the prompt so you must answer before anything is handed to you.
Principle evidence
Strength of the underlying learning idea, not a claim about AI products.
The underlying learning idea is rated strong. Answering escalating questions so you explain in your own words is a well-supported generative practice. Explaining beats re-reading when study time is held constant.
AI delivery evidence
Whether an AI tutor delivers this method well is a separate question.
Whether an AI interviewer delivers that benefit as well as a human tutor or a worksheet is a separate question. Treat AI delivery as speculative unless a tutoring study with this method is named, and none is claimed here.