The refusal-to-answer request

Instruct it, up front, that it is not allowed to give you answers, only questions.

When to use this: While you practise
Learning impact: Strong change
Learning evidence: Credible with caveats
Where you are with the topic: Building understanding
How often to use it: Build and keep as a habit

When to use it

You notice you are collecting explanations instead of producing them.
A Socratic or practice session keeps drifting into lectures.
You want the AI to interview you, not teach at you.

What to ask AI

Type something like: “For this entire session: you may not give me answers, explanations, or solutions. You may only ask me questions.”

If you ask a direct question, it must respond with a question that helps you find it yourself.

Keep the hard part on your side: you retrieve and explain. Override the rule and the method ends.

How it helps

Helpfulness is the default. Left alone, the model explains.

You let it, because being told is easier than working.

A refusal-to-answer request makes withholding the rule instead of a mood.

Question-only tutoring pushes generative learning: you construct answers rather than consume them.

The method fails the same way every time: drift back into explaining within a few turns, and you accept the gift.

How to run it

State the rule at the top, before any content question.
Name the topic narrowly enough that questions can bite.
Answer every question before asking for a hint. Hints should still be questions.
When it slips into explaining, paste the rule again and discard the explanation.
Stop when you hit a question you cannot answer. That gap is the destination.
Close the loop in your notes or textbook, then re-enter with the same rule.

Example exchange

You

For this whole session, don’t give me answers or explanations, only questions. The topic is why voltage drops across a resistor in a simple series circuit.

In your own words, what is being “used up” as current moves through the resistor: charge, energy, or something else?

You

Energy? The charges keep going, so maybe electrical potential energy.

What would you measure before and after the resistor to see that change? Don’t look it up yet.

Copyable prompt

For this entire session: you may not give me answers, explanations, or
solutions. You may only ask me questions. If I ask you a direct question,
respond with a question that helps me find it myself. If I get stuck for a long
time, you may give me the smallest possible hint, but never the answer. Confirm
you understand, then begin.

The Tell

Here is how you know this method has flipped: you finish the session having been told several things.

Either you overrode the rule or it drifted and you let it. Both feel identical afterward.

That is why the rule has to be explicit at the start, and enforced when it slips.

Principle evidence

Strength of the underlying learning idea, not a claim about AI products.

The underlying learning idea is rated moderate. Socratic withholding and question-only tutoring push generative learning: the learner constructs answers rather than consuming them. Configuring the AI to ask-only is a delivery wrapper on that principle. Principle moderate for withholding-as-tutoring; sustained AI compliance with ask-only rules is uneven.

AI delivery evidence

Whether an AI tutor delivers this method well is a separate question.

Models drift into helpful explanations. No claim here that ask-only mode stays stable without your enforcement. Delivery is speculative; your refusal to accept answers is what keeps the method alive.