Self-Explanation Is More Than Putting It in Your Own Words

“Explain it in your own words” sounds active. It can still be shallow.

Imagine a spreadsheet tutorial that says:

A paraphrase might be:

The wording changed. The reasoning did not become visible.

A self-explanation goes further:

Now the explanation contains a relationship, a reason, and a condition.

Self-explanation is the attempt to generate an inference about how parts connect: why a step follows, how an example reflects a principle, or under what conditions a choice works. It can be written, spoken, sketched, or silent.

The value comes from the reasoning you generate—not from sounding polished.

What the Self-Explanation Effect Research Actually Shows

The foundational study is smaller and more specific than its fame suggests.

In 1989, Michelene Chi and her colleagues studied how learners worked through solved physics problems. The stronger learners did more than describe the examples. They connected solution steps to principles, expanded the conditions behind those steps, and noticed where their own understanding broke.

They also became less dependent on the exact examples they had studied.

That study helped define the self-explanation effect, but it was a small observational study. It showed a meaningful pattern. By itself, it did not prove that prompting any learner to explain any material would cause the same result.

Later experiments tested narrower versions of the idea.

In a randomized study of 56 bank apprentices, Alexander Renkl and his colleagues used worked examples involving interest calculations. Prompts designed to elicit stronger explanations supported transferable knowledge. The clearest near-transfer benefit appeared among learners with lower prior knowledge.

When a solved example is carrying you through unfamiliar territory, asking why each decision works can turn the example from something you copy into something you can reuse.

The wider evidence is promising too. A 2018 meta-analysis combined 69 effect sizes from 64 research reports. The random-effects average was Hedges’ g = 0.55—a moderate result across a mixed evidence base.

But “mixed” matters.

Those reports included different subjects, ages, prompts, activities, and comparison groups. An average across varied conditions does not mean that every version of self-explanation works equally well.

The useful conclusion is not “explaining always wins.”

It is that well-aimed explanation prompts can improve learning, especially when they make you build connections the example did not state for you.

A New Result Draws an Important Boundary

The cleanest correction comes from a 2026 preregistered experiment.

The researchers assigned 208 adults to self-explain or reread information about fictitious species. They kept time on task equal and tested factual knowledge immediately and again two weeks later.

The self-explanation condition did not outperform equal-time rereading on factual retrieval at either test.

Within the explanation group, producing more correct inferences was associated with better performance. Yet being assigned to self-explain did not cause a better factual-learning result than rereading in that task.

That sharpens the job description.

The study deliberately used unfamiliar factual material and minimized prior knowledge. A worked solution, a causal model, and a list of facts are not the same target.

If I need to remember a product code, a vocabulary definition, or a sequence of labels, I do not need to manufacture a grand explanation for each item. I need to see whether I can produce the answer without looking. Our guide to active recall versus rereading covers that job directly.

If I need to understand why a formula changes, why a strategy fits one case but not another, or why a process failed, self-explanation has something different to work on.

It can expose the missing relationship.

The Explain the Decision Test

I use five questions to turn a worked example into a real understanding check.

You can apply them to code, accounting, design, statistics, writing, software, or any skill where an example contains decisions.

Do not wait until the end of a 40-minute lesson. Stop after one meaningful step.

1. What changed in this step?

Describe the move precisely.

Not “then they fixed it.”

Try:

• →the function changed from returning a list to returning one matching item;
• →the designer increased the spacing between groups, not between every element;
• →the calculation switched from total revenue to revenue per customer; or
• →the editor removed a claim because the source supported correlation, not causation.

If you cannot identify what changed, the example is still a blur.

2. Why was this the right move here?

Name the problem the step solves.

This question separates a decision from a ritual.

“Because the tutorial did it” is not an explanation. “Because the value must stay constant while the row reference changes” is.

If your answer contains only a restatement, ask again:

The consequence often reveals the reason.

3. What rule or relationship makes it work?

Move one level above the example.

The exact numbers, colors, variable names, or sentences belong to this case. The principle is what you want to carry forward.

For the spreadsheet example, the principle is not “press F4.” It is that one reference must remain fixed while another changes across copied formulas.

For an argument, the principle might be that the conclusion cannot be stronger than the evidence.

For a design, it might be that spacing signals grouping.

This is where self-explanation becomes reusable.

4. When would this move be wrong?

Every useful rule has conditions.

Ask what would have to change before you chose a different step.

• →What if every row used a different tax rate?
• →What if the function should return all matches?
• →What if the study randomly assigned participants?
• →What if the visual elements belonged to one group instead of two?

This question protects you from turning one example into a rigid recipe.

It also makes uncertainty visible. If you cannot name the boundary, mark it as a question instead of improvising an answer.

5. Can I use the rule after the surface details change?

Now remove the example.

Change one important detail and make a new attempt. Use a different dataset. Rewrite the code with new names. Diagnose a fresh paragraph. Apply the same design principle to another screen.

This is the part most “explain it simply” advice leaves out.

An explanation can sound complete and still fail to guide action.

Transfer is the harder test.

If the new attempt works, you have evidence that the principle traveled. If it fails, return to the smallest missing decision, check the source, and try again.

That repair cycle fits the four-part learning lab: learn, retrieve, correct, and return. Self-explanation belongs inside the loop. It does not replace the loop.

Check the Explanation Before You Trust It

Self-generated does not mean self-correcting.

I can build a perfectly coherent explanation around a false assumption. The explanation may even increase my confidence because I supplied the missing story myself.

So every self-explanation needs a check.

Use the answer key, documentation, source paper, instructor demonstration, test output, or another authoritative reference. Compare the specific rule and its conditions—not just the final answer.

When the explanation is wrong, do not merely copy the correct wording.

Write the correction as a contrast:

That turns an error into a boundary you can recognize later.

The latest evidence reinforces this point. A 2026 replication involving 115 university students found evidence that explanation quality could help account for worked-example benefits, but the pathway varied with prior knowledge and working-memory capacity. The authors were explicit: the mechanism is not universal.

More explanation is not the target.

Better, checked reasoning is.

When Self-Explanation Earns the Time

Use the Explain the Decision Test when the material contains:

• →a worked solution with multiple steps;
• →a cause-and-effect chain;
• →a choice among strategies;
• →a rule that changes with conditions;
• →a process you must diagnose; or
• →an example you will need to adapt.

It is especially valuable when everything feels clear while the example is visible. That feeling is exactly what needs a stronger test.

Use a different tool when the goal is different.

For bare factual retention, close the source and retrieve. For physical execution, practice the movement and use feedback. For creative judgment, compare real outputs against criteria. For a completely unfamiliar, high-complexity task, get a clear explanation or a simpler worked example before demanding that you generate one yourself.

An evidence review by Bethany Rittle-Johnson and Abbey Loehr describes the same caution: explanation prompts direct attention. They can help one outcome while pulling effort away from another, and an alternative technique may be stronger for a different job.

The strategy has to match the test you care about.

What This Evidence Does Not Prove

The research does not establish one perfect script for every adult learner, subject, or level of expertise.

It does not show that talking longer produces better understanding.

It does not guarantee that an explanation is correct, durable, or transferable without feedback and another attempt.

It does not mean that self-explanation replaces retrieval practice, spaced return, direct instruction, worked examples, or deliberate practice.

And it does not turn the ability to sound simple into proof of mastery.

The better standard is behavioral:

Can you justify the decision, identify its boundary, and use the principle when the example changes?

Stop Agreeing with the Example

The next time a tutorial feels obvious, pause before the next step appears.

Ask what changed.

Ask why it was right here.

Name the rule. Find the condition. Change the problem.

Then check your reasoning against something that can correct you.

You may discover that you understood more than you thought.

You may discover that one confident sentence was holding up the entire illusion.

Today, choose one worked example and run the Explain the Decision Test before you copy the next move.

Do not ask, “Does this look familiar?”

Ask, “Can this explanation make the next decision without help?”

That is where understanding starts to become usable.