Discovery-first guided inquiry

Be led to an idea by evidence and questions until you produce it yourself, rather than being told it and then shown why it's true.

When to use this: First time through the material
Learning impact: Strong change
Learning evidence: Credible with caveats
Where you are with the topic: Building understanding
How often to use it: Worth repeating often

When to use it

You are meeting a central idea for the first time and want to arrive at it yourself.
The topic is known for stubborn misconceptions that survive being told the right answer.
You have time for a slower path that sticks better than a quick definition.

What to ask AI

Type something like: “I want to work out [target idea] for myself. Do not tell me it and do not define it.”

Ask for one concrete situation at a time, then a question about what happens.

Keep answering before you ask for the name of the idea.

How it helps

Being told the right answer often leaves the wrong intuition intact.

You can state the correct definition and still reason from the incorrect model five minutes later.

Discovery-first guided inquiry flips the order: the AI supplies cases, not conclusions.

Each case is chosen so that reasoning about it produces a piece of the idea.

You assemble the concept; the questioner sequences the evidence.

An idea you produced is retained better than one you only received.

The caveat matters: unguided discovery (wander with almost no support) has a poor research record.

What works is the guided version, sequenced cases, aimed at a specific idea, with feedback on the path.

How to run it

Tell the AI the target idea secretly. It needs to know where it is leading; you must not.
Demand cases rather than explanations: situations to reason about, not statements to agree with.
Commit to an answer at every step. Passive reading of a transcript does nothing.
Expect a case your current idea cannot handle. The stall is where the next piece comes from.
State the idea in your own words at the end, then check the textbook version.
Reserve this for central, misconception-prone ideas. It is slower than being told, and that cost should buy something important.

Example exchange

On an island, some beetles are green and some are brown. The ground is brown. Birds eat beetles. What do you think happens over many generations?

You

The green ones get eaten more because you can see them against the ground.

So after many generations, what does the beetle population look like?

You

Mostly brown.

Do the individual green beetles turn brown to survive?

You

No, they die. The population changes, not each beetle.

Copyable prompt

I want to work out [TARGET IDEA] for myself. Don't tell me it and don't define
anything. Give me one concrete situation at a time and ask me what I think
happens. Choose each next situation so that my answer to the last one reveals
what I still haven't got. If I state the idea, tell me whether it's complete.

The Tell

If you were led somewhere and cannot say what evidence took you there, you were told, slowly.

If the AI named the concept before you could, the discovery never happened.

Demand cases until you can state the idea without hearing its name first.

Principle evidence

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

The underlying learning idea is rated moderate. Well-scaffolded guided inquiry can work; pure discovery with minimal guidance often does not. The strength is in guidance plus generation, not in leaving you to wander.

AI delivery evidence

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

How well an AI chooses the next case and withholds the punchline is delivery quality, not principle strength. Treat AI inquiry guidance as speculative here.