Learn Better With AI

Methods · Generative practice & retrieval

B7

Immediate targeted feedback

Attempt first, then get a correction aimed at your specific mistake while you still remember why you answered that way.

When to use this
While you practise
Learning impact
Highest-leverage move
Learning evidence
Strong research tradition
Where you are with the topic
Strengthening memory
How often to use it
Worth repeating often
What AI is doing (1)Evaluate your work
What AI is doing (2)Reveal what you are missing

When to use it

  • Homework comes back a week later with a mark and you cannot remember your reasoning.
  • You want to know the first step that went wrong, not a full correct solution dumped on you.
  • You keep making the same error and generic “review chapter 3” advice is not helping.
  • You have written an attempt and need a diagnosis while the thinking is still warm.

What to ask AI

Paste the problem and your full attempt. Say: “Do not give the correct solution yet. Tell me (1) the first step I went wrong, (2) what I seem to believe that made me do it, and (3) one question that would help me see it.”

Ask for the first error only when you made a chain of mistakes.

Keep the solution hidden until you have repaired the first break.

How it helps

Feedback works when you committed to an answer, the correction arrives while you still remember why, and it addresses your actual error.

A mark a week later lands on nothing. You have already forgotten the path you took.

“You were wrong” is grading. “You treated the minus sign as belonging to the term” is teaching.

AI removes the waiting-list problem of one teacher and thirty students, if you show your working.

How to run it

  1. Commit in writing, including the step you guessed.
  2. Show working, not only a final answer.
  3. Ask for the first error, the likely belief behind it, and one probing question.
  4. Answer that probe yourself before requesting the next hint.
  5. Only then ask to check a repaired attempt, still prefer hints over a full rewrite.

Example exchange

You

Problem: differentiate y = (3x−2)⁵. My attempt is 5(3x−2)⁴. Please do not give the full solution. Tell me the first wrong step, what I seem to believe, and ask one question.

AI

First issue: the chain-rule factor for the inner function is missing. You seem to believe the outer power rule finishes the job alone. Question: what is the derivative of (3x−2) with respect to x?

You

It is 3. So I needed to multiply by 3.

AI

Yes. Rewrite that line with the factor included, then I will confirm, still no full model solution unless you ask.

Copyable prompt

Here is a problem and my attempt. Do NOT give me the correct solution yet.
Tell me only: (1) the first step where I went wrong, (2) what I appear to
believe that made me do it, and (3) one question that would help me see it.

PROBLEM: [paste]
MY ATTEMPT: [paste everything, including uncertain steps]

The Tell

You will know the method has flipped when you paste a problem and ask for the right answer without an attempt, or when the AI returns a full perfect solution you only recognise.

Then nothing was targeted, and nothing was yours.

Force the three-part diagnosis on your working before any correct solution appears.

Principle evidence

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

Immediate, specific feedback after an attempt is foundational in skill learning and formative assessment. Targeting the error, not only the grade, is the load-bearing move. Explaining feedback beats right/wrong marks alone.

AI delivery evidence

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

The principle is strong; AI diagnostic accuracy is not proven equal to a good human tutor. An AI can be confidently wrong about whether you were wrong. Verify surprising diagnoses against a trusted source.

Related methods