MindOS Learning Manual: Learning-by-Teaching State | Preparing to Teach Can Change How You Learn

Wait, What?

You can sometimes learn more because you expect to teach someone else.

That does not mean every student should become the teacher.

It means the prospect of teaching can change the learner’s study operation. The learner may select more carefully, organise ideas more coherently, retrieve more actively, notice gaps sooner and generate explanations for another mind rather than merely reread for themselves.

Then the act of actually teaching can add another round of retrieval, explanation and monitoring.

Quick Answer

Learning-by-Teaching State is the learner operation of studying with the intention to teach, then teaching or explaining the material to another person—or a realistic imagined audience—and finally checking what the learner themselves can still retrieve, explain and transfer.

Research does not support the slogan “teaching is always the best way to learn”. Effects vary. But recent meta-analytic work suggests an important pattern: preparing to teach and actually teaching can make separate contributions, and teaching appears more useful when the learner prepared with teaching expectancy than when teaching is imposed after ordinary study.

MindOS therefore treats learning by teaching as a structured generative operation, not a motivational trick.

Owned Learning Operation

LEARNING-BY-TEACHING STATE = prepare to teach → organise for another mind → retrieve and explain → detect gaps/questions → repair → teach again or reconstruct → test independent retention and transfer.

This is narrower than Social Learning State, which owns peer interaction becoming individual capability. It is also narrower than Explanation State, which owns the learner’s ability to explain relationships. Learning by teaching specifically adds an instructional audience and teaching intention.

Teaching Changes the Study Question

When students study for themselves, the internal question can quietly become:

Does this look familiar enough?

When they know they must teach, the question often becomes more demanding:

What is the main idea?
What does the other person need first?
How do I explain the relationship?
Which example would make this clearer?
Where could they misunderstand me?
Can I answer a question if they interrupt?

Those are generative decisions. They require selection, organisation, retrieval, explanation and anticipation.

But the learner can also spend all their effort on performance: slides, jokes, delivery, appearance, recording quality or sounding confident. That is why the operation needs a firewall.

The Performance-Theatre Trap

A learner can produce a polished five-minute teaching video and learn very little.

Possible reasons include:

the script was copied from the source;
the learner read continuously instead of retrieving;
visual production consumed most of the effort;
the explanation remained superficial;
the learner avoided areas they did not understand;
AI generated the teaching script;
the audience never asked questions;
the learner was assessed on presentation quality rather than knowledge.

MindOS does not count the teaching artifact as proof. The receipt is what the learner can do afterward with less support.

Four Learning-by-Teaching States

1. Teaching Expectancy Only

The learner studies while expecting to teach but never teaches.

This can alter study behaviour, but the later explanatory retrieval phase never occurs.

2. Teaching Without Teaching Expectancy

The learner studies normally and is suddenly told to teach.

Meta-analytic evidence suggests this can be much less effective than preparing specifically to teach.

3. Prepared Teaching

The learner studies knowing they must teach, organises the material for another mind, then actually explains it.

This is the core MindOS state.

4. Interactive Teaching

The other learner asks questions, challenges explanations or reveals misunderstanding. This can create additional learning opportunities, but it also adds social and task complexity. It should not be confused with the simpler non-interactive teaching effect studied in many experiments.

The MindOS Learning-by-Teaching Protocol

Step 1 — Define the Teaching Target

Teach one bounded idea, not an entire subject.

Examples:

why multiplying a negative by a negative gives a positive;
how diffusion differs from osmosis;
how an inference differs from a literal retrieval answer;
when to use completing the square;
how evidence supports a claim in an argument.

Step 2 — Prepare for an Audience, Not for a Script

Ask:

What must the other learner already know?
What is the central relationship?
Which example will reveal it?
Which misconception is likely?
Which question would expose weak understanding?

Notes may be used during preparation, but the final teaching should gradually require more reconstruction and less reading.

Step 3 — Teach Without Reading the Source

Close the textbook or notes as much as the learner’s level allows. Teaching should trigger retrieval, not become oral copying.

Step 4 — Force an Example and a Boundary

Do not stop at definition-level teaching. Require:

one correct example;
one near non-example;
one explanation of why the difference matters.

Step 5 — Invite a Question

If another person is present, ask them for the part that remains least clear. If no person is present, generate a hostile-but-fair question: “What would someone ask if they did not believe my explanation?”

Step 6 — Repair the Gap

Return to the source only where the explanation broke. Then teach that part again from memory.

Step 7 — Separate From the Audience

Now test the learner alone. Can they retrieve, explain and use the knowledge independently?

Worked Examples Across Subjects

English: teach a younger student what makes evidence relevant rather than merely present. The learner must select examples, explain the connection between evidence and claim, then answer a new question alone afterward.

Mathematics: teach why cross multiplication works instead of presenting it as a magic procedure. If the learner cannot explain the preserved proportional relationship, the teaching exposes procedural knowledge without conceptual ownership.

Science: teach the difference between mass and weight. Require one Earth example, one Moon example and one near misconception. Then remove the audience and test the learner with a new scenario.

Competing Explanations When Teaching Helps

Preparing to teach may improve selection and organisation.
Teaching may function partly as retrieval practice.
Explaining may generate elaboration and integration.
Audience awareness may improve clarity and sequencing.
Questions may expose knowledge gaps.
Social responsibility may increase effort or attention.
The learner may simply spend more productive time with the material.

Research has not reduced learning by teaching to one universally accepted mechanism. MindOS therefore focuses on the observable operations rather than claiming one hidden cause.

How Do We Know?

A 2024 meta-analysis by Kobayashi examined the interaction between preparing to teach and actually teaching. The pattern is particularly useful for practice: teaching after preparing with teaching expectancy showed a medium learning benefit relative to ordinary study, while teaching after study that did not include teaching expectancy did not show the same advantage. Actually teaching after preparing to teach also added a small-to-medium benefit beyond preparation alone.

A 2022 meta-analysis of students generating teaching materials found a small overall advantage, but the result depended strongly on what students created and what the comparison condition was. Visual and audio-visual teaching materials tended to outperform text-only products, and the advantage was clearer against weak control conditions than against other already-effective learning activities.

A 2025 large-scale classroom analysis also emphasised heterogeneity: non-interactive teaching can help, but effects vary across learners, contexts and study designs. That is an important correction to popular claims that “the best way to learn is always to teach”.

What This Evidence Does Not Prove

It does not prove that teaching always beats retrieval practice, self-explanation or worked examples.
It does not prove that every learner benefits equally.
It does not prove that polished teaching materials indicate strong learning.
It does not prove that an audience must be real or interactive for any benefit to occur.
It does not prove that teaching expectancy alone is enough for durable learning.

The safe inference is narrower: preparing to teach can change how learners study, and actually teaching can add further generative work, but the benefit depends on how much real selection, retrieval, explanation and repair the learner performs.

AI Boundary: Do Not Outsource the Teaching

AI can help generate questions, simulate a novice audience or challenge an explanation. But if AI writes the lesson, selects the examples, explains the misconceptions and answers the questions, the learner may perform the teaching while the tool performs the learning operations.

A safer AI-assisted sequence is:

learner prepares first;
learner teaches from their own model;
AI asks questions or flags unclear points;
learner repairs the explanation;
AI closes;
learner reconstructs and solves independently.

Better teaching output is not automatic evidence of better learner capability.

When Learning by Teaching Is the Wrong Tool

When prerequisite understanding is too weak to teach accurately.
When the learner needs a worked example before generating an explanation.
When teaching performance anxiety consumes the cognitive work.
When the task is simple enough that teaching adds unnecessary overhead.
When the learner repeatedly teaches misconceptions without corrective feedback.
When the real weak link is retrieval speed, not organisation or explanation.

Scaffold Fade

Stage 1: learner teaches with a concept list and prompt questions.
Stage 2: learner teaches with a small outline.
Stage 3: learner teaches with only the title or target question.
Stage 4: learner answers audience questions without the source.
Stage 5: learner stops teaching and proves capability on a new independent task.

The teaching scaffold succeeds when it can disappear.

Immediate, Delayed and Transfer Checks

Immediate: can the learner teach without reading?
Question test: can the learner answer a reasonable challenge?
Delayed: can the learner reconstruct the topic later without the teaching script?
Transfer: can the learner apply the principle to a new case?
Independence: can the learner perform when there is no audience, no presentation and no teaching role?

Teaching Guide for Parents, Tutors and Teachers

Ask the learner to teach something small enough to reveal structure.

Instead of “teach me the whole chapter,” try:

“Teach me why this step is allowed.”
“Teach me the difference between these two concepts.”
“Teach me using one example I have not seen before.”
“What question should I ask if I think your explanation is too simple?”
“Now stop teaching and solve this new problem yourself.”

The final question protects the learner job. Teaching is a method. Independent capability is the outcome.

MindOS Direction

If the learner cannot explain the relationship while teaching: route to Explanation State.

If the learner teaches confidently but knowledge later disappears: route to Retrieval State, Spacing and Successive Relearning.

If the teaching relies on one familiar example: use Example Generation, Practice Variability and Transfer State.

If group interaction allows one learner to carry the thinking: use Social Learning State.

MindOS rule: teaching is educationally valuable when preparing for another mind forces the learner to organise, retrieve, explain and repair—and when those gains remain after the audience disappears.