Methods · Metacognition & self-regulation · Short guide

Self-explanation prompting

Explain each step to yourself while studying, turning passive reading into generative monitoring that catches the first fuzzy step.

When to use this: First time through the material
Learning impact: Highest-leverage move
Learning evidence: Strong research tradition
Where you are with the topic: Building understanding
How often to use it: Worth repeating often

When to use it

You are working through a worked example or dense paragraph and keep nodding along without checking each step.
Reading feels fluent and you want generative monitoring that catches the first fuzzy step.
You finished an example yesterday and cannot justify a step you handwaved.

What to ask AI

Put it in self-explanation mode: reveal one step at a time; wait for your explanation of what it does and why it is allowed; probe if vague; do not explain for you.

When you fail, ask it to stop and name the gap, then you consult notes, not the AI lecture.

Keep AI as nag and gap-spotter, not as the explainer.

How it helps

Reading creates fluency. Self-explanation creates the retrieval and inference work that reading skips.

This is during study: after each step or sentence, you produce a why/how before continuing, not a later teach-back, and not a Socratic interview.

The method is easy to fake. Mouthing “makes sense” is not an explanation. Force a concrete utterance: what this step does, why it is legal, what would break if omitted.

The first failure is the study target. Fix that node, finish the example, then re-solve a twin without notes.

How to run it

Take one worked example or dense paragraph.
After each step or sentence, pause and write or say one or two lines: what it does, why it is allowed.
Instruct AI to probe empty or circular explanations only, not to teach ahead.
Mark the first failure and study that node from your materials.
Finish the example, then re-solve a twin problem cold.

Example exchange

You

Can we do self-explanation on this completing-the-square example? Show one step at a time and wait for me. Don't explain it for me.

Step one: x² + 6x = 7. What does this setup buy you, and why is moving the constant allowed here?

You

We're isolating the quadratic bits so we can make a square? I'm not sure why the 7 can move.

One probe only: what property of equality lets you move a term across the equals sign? Say it in one clear sentence before we go to step two.

You

If I add or subtract the same thing on both sides, the equality still holds, so I can subtract 7 from both sides.

That's enough. Here's step two when you're ready, same rules.

Copyable prompt

Self-explanation mode on this worked example / text: [ ].
Reveal or consider ONE step at a time. Wait for my explanation of what it does
and why it's allowed. If I'm vague, ask one probe. Do not explain for me. When
I fail, stop and name the gap; wait for me to consult my notes/textbook.

The Tell

If you finished the example and cannot justify the step you handwaved, the AI (or your pace) carried you past the learning.

The pause at the fuzzy step is the method. Skipping it because it feels slow is how fluency wins again.

Principle evidence

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

Strong evidence for self-explanation while studying worked examples and text: generating why/how at each step improves monitoring and learning compared with passive reading. The principle is about your explanations, not about receiving better lectures.

AI delivery evidence

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

AI prompts can enforce the pause-and-explain rhythm. If the model starts explaining “to keep you moving,” delivery has failed. Efficacy of AI prompting versus a paper worksheet is still speculative; the cognitive act is yours.