Misconception elicitation

Answer questions designed to catch the specific wrong models people hold in your subject, and find out which ones you're carrying.

When to use this: When something feels wrong
Learning impact: Strong change
Learning evidence: Strong research tradition
Where you are with the topic: Checking readiness
How often to use it: Worth repeating often

When to use it

You keep getting “easy” questions right and still fail novel ones that look similar.
You can recite the definition and reason from its opposite under time pressure.
A teacher mentions a “common misconception” and you are not sure whether you hold it.
Homework scores look fine, but explanations never feel solid.

What to ask AI

First ask what the documented misconceptions are for the topic, and what wrong model produces each.

Then ask for eight questions that give different answers under the correct model versus each wrong one, with targets hidden.

Answer fast (about ten seconds each), then ask which wrong model each wrong answer fits.

Ask for one concrete case that kills the model you are carrying, not a lecture of the right answer.

How it helps

A gap is empty. You know you do not know; the remedy is to learn it.

A misconception is occupied. You have a confident, coherent, wrong model that produces right answers on practised questions and wrong ones elsewhere.

From the inside it feels exactly like knowledge, so you will not flag it yourself.

Ordinary questions do not discriminate. You need items where the misconception gives one specific wrong answer and the correct model gives another.

The remedy is not more explanation beside an untouched wrong model. Activate it, show it fail on a case you care about, then replace it.

How to run it

Ask for the known misconceptions in the topic before any testing.
Ask for discriminating questions with the targets hidden.
Answer quickly without carefully reconstructing. The fast answer is the model you actually run.
Have wrong answers diagnosed by model: “that is what someone gets if they believe X.”
Demand the killing case, a concrete situation where your model is obviously wrong and you can check it.
Re-test in a fortnight. Misconceptions return under time pressure and fatigue.

Example exchange

You

I want to catch Newton's first law misconceptions. Quiz me with discriminating questions, but don't tip which misconception each one targets.

Eight quick items. Answer each from gut instinct, under ten seconds if you can.

You

1B, 2A, 3B, 4A, 5B, 6A, 7B, 8A.

Your pattern on items 2, 5, and 7 fits the “force in the direction of motion” model. Killing case: a hockey puck sliding on ice after the stick leaves, what force keeps it moving?

You

None? It just keeps going because nothing stops it?

Right. Check that gut against the law before you revise anything else.

Copyable prompt

Give me eight questions designed to detect documented misconceptions about [TOPIC]:
each one should give a different answer depending on which model I'm using. Don't
tell me which misconception each targets. I'll answer fast, without working it out.
After my answers, tell me which wrong MODEL each wrong answer is consistent with,
then give one concrete case that kills my model.

The Tell

You get everything right, slowly.

Speed up: the fast answer is the one the exam gets.

If careful reasoning always rescues you in practice, you have not seen the model you run under pressure.

Principle evidence

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

The underlying learning idea is rated strong. Items designed to expose common wrong models come from conceptual-change assessment design (diagnostic distractors), with a strong tradition in science education. Principle strength is high for validated misconception items.

AI delivery evidence

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

AI-written discriminating items need validation. Treat them as practice diagnostics that name candidate models, not as official inventory scores.