Audience-specific explanation

Explain the same idea to a novice, a peer, and a hostile expert in turn, adjusting grain and warrants without changing the truth-conditions of the claim.

When to use this: While you practise
Learning impact: Strong change
Learning evidence: Credible with caveats
Where you are with the topic: Applying in new contexts
How often to use it: Worth repeating often

When to use it

You can explain an idea only one way, usually the way your notes are written.
You need to teach, present, or write for audiences that do not share your context.
You want a check that you understand, not only that you can recite.

What to ask AI

Type something like: “I will explain gradient descent three times: to a novice, to a peer, and to a hostile expert. Critique each for audience fit. Do not rewrite for me; make me revise.”

You produce all three; AI critiques fit.

How it helps

Audience-specific explanation is generative learning: adjusting grain, jargon, and objections for who is listening.

If you can only say it one way, you may only recognise it one way.

How to run it

Pick one idea and three audiences.
Explain to each without copying the previous version.
Take critique on audience fit, not generic “clarity.”
Revise the weakest audience yourself.
Note which version exposed a gap in your understanding.

Example exchange

You

Idea: statistical significance. I need versions for a novice journalist, a methods peer, and a hostile domain expert.

Your novice draft used “proves.” That audience leaves with a falsehood. Revise without “prove.”

You

“Suggests the result is unlikely under a no-effect model, given the assumptions…”

Better. The hostile expert will ask which assumptions,name two now, before we polish tone.

Copyable prompt

Idea: [precise sentence].
I will explain it three times: novice, peer, hostile expert. After each, only
tell me: (1) where I got vague, (2) where I may have become false, (3) one
question that audience would ask next. Do not rewrite for me.

The Tell

You know this method has flipped when all three “audiences” get the same paragraph with different greetings.

That is find-and-replace, not audience control.

If a hostile expert version never changes your claim, you did not meet them.

Principle evidence

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

The underlying learning idea is rated moderate. Audience-specific explanation is generative learning / teaching-as-test plus register control (related to explaining the idea in your own words and Feynman-style explain-back).

AI delivery evidence

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

Principle moderate; AI audience critique speculative.