Fading / graduated withdrawal

Design a short sequence that starts with full support and ends with you doing the work alone, and advance only when you've earned the next rung, instead of staying on crutches that feel like studying.

When to use this: While you practise
Learning impact: Strong change
Learning evidence: Strong research tradition
Where you are with the topic: Building understanding
How often to use it: Build and keep as a habit

When to use it

You have twenty minutes tonight and need real practice, not more highlighting.
Homework is done, but you still want one honest attempt before bed.

What to ask AI

Type something like: “I want a fading sequence for this skill: …. Give me:”.

Tell it your current level and ask it to stay one step harder than comfortable.

Keep the hard part on your side: you retrieve, attempt, explain, or decide before you ask for help.

How it helps

Contingent hinting (Contingent hinting) answers the emergency: you're stuck now, and you need the smallest help that unsticks you.

Fading answers the longer problem: help that never leaves becomes the skill. You who always studies with worked examples open, always asks for the first step, always has the AI check mid-solution, can look productive for weeks and still freeze on the first unaided exam question.

Nothing went wrong in any single session.

What went wrong is the schedule of support, it never changed.

The mechanism is scaffolding with withdrawal.

Temporary support that is removed as competence grows.

The research name is fading; the everyday failure mode is "I still need the training wheels" said three months later.

AI makes support so cheap and pleasant that withdrawal has to be deliberate, or it never happens.

How to run it

Pick one skill, narrowly. "Quadratic equations" is too wide. "Completing the square when a ≠ 1" is a skill you can fade across.
Next: stage 1, study, don't copy. Ask for a full worked example with every decision point labelled. Close it. Reconstruct from memory. Open and compare.
Next: stage 2, completion. Ask for a near-twin with the last third blank (Completion problems). Finish it. Only then see the model answer.
Next: stage 3, hints. New problems, Contingent hinting ladder only. Log the rung you needed.
Next: stage 4, cold. Timed or closed-tab. No hints until you've submitted an attempt.
Do not advance on vibes. Advance on the criterion in the table. Feeling ready is the least reliable signal you have.

Example exchange

You

I want support that fades for related-rates problems. Start with one fully worked example. Wait for me before the next rung.

Here's a fully worked tank-draining example with every decision labelled. When you've studied it, say ready and I'll give a near-twin with the last third blank.

You

Ready. I filled most of the blank, but I stalled choosing which variable to differentiate with respect to time.

That's the decision this sequence is training. Next comes a fresh problem, hint only if you ask.

Copyable prompt

I want a fading sequence for this skill: [SKILL]. Give me:
(1) one fully worked example with every decision point labelled,
(2) one near-twin with the last third blank,
(3) three fresh problems for hint-only practice,
(4) two cold unaided problems.
Do not give solutions for (3) or (4) unless I ask for a contingent hint under
A2 rules. Wait for my attempts.

The Tell

Here is how you know this method has flipped: you can do the skill with the chat open and cannot start the same skill with the chat closed. Support never withdrew; the method never ran.

In that moment the AI (or the schedule, or the story) did the thinking, and you only recognised a finished product.

Tighten the prompt for fading / graduated withdrawal so you attempt, decide, or retrieve before anything is handed to you.

Principle evidence

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

The underlying learning idea is rated strong. Graduated withdrawal is the operational half of scaffolding (Wood et al.): full model → partial → prompts → unaided, advancing on success. Instructional-design work on fading in worked examples and guided practice treats the same sequence. Principle evidence for fading as a tutoring pattern is strong; AI-managed stage advancement remains speculative.

AI delivery evidence

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

The learning principle may be strong, but that does not prove an AI session delivers it well. Treat AI delivery as speculative unless a study of tutoring with this method is named, and none is claimed here.