How Writing to Learn Works | Why Putting an Idea Into Words Can Change What You Understand

A student says, “I understand it.”

The teacher replies, “Write three sentences explaining why.”

The student writes one sentence, stops, crosses out two words, stares at the page and then says something more useful: “Actually, I understand the first part, but I cannot explain how the second part follows.”

Nothing new has been taught in those thirty seconds. Yet the learner now knows more accurately what is known.

That is one reason writing can function as a learning tool rather than merely as a way to display learning. Putting an idea into words forces selection. The writer must decide what matters, organise relations, retrieve terminology, make connections explicit and discover where the explanation breaks. A vague feeling of familiarity can survive silently in the mind. It is harder to hide once the learner has to build a sentence that another person could follow.

This does not mean “more writing always means more learning.” A student can copy notes for an hour with almost no conceptual gain. A poorly designed writing task can consume time that would be better spent on explanation, practice or feedback. Writing can also overload a learner who is simultaneously struggling with handwriting, spelling, language, subject knowledge and complex reasoning.

The useful question is therefore not, “Should students write more?” It is: What kind of writing makes the learner do the thinking the subject requires?

That is the mechanism this page owns.

The 50-second answer

Writing to learn works when writing is used as a cognitive operation: the learner retrieves, selects, organises, explains, compares, justifies or connects subject knowledge in order to produce a meaningful account.

A practical mechanism looks like this:

1. The learner receives a precise thinking prompt. Not merely “write about the topic,” but explain, compare, predict, justify or connect.
2. Knowledge must be retrieved. The learner cannot rely only on recognition if notes are closed or partially restricted.
3. Ideas must be selected and organised. Writing requires decisions about sequence and relation.
4. Implicit gaps become visible. The learner discovers where a causal link, definition or evidence connection is missing.
5. Language forces precision. Terms must be chosen and relations made explicit enough for a reader to follow.
6. The product creates evidence for feedback. Teacher, peer or learner can inspect the reasoning rather than guess at it.
7. Revision can change the knowledge structure. Correcting the writing can also correct the understanding.
8. Transfer is tested on a fresh prompt. The learner should eventually explain the idea without relying on the original wording or frame.

The purpose is not always polished prose. Sometimes the best writing-to-learn task is five rough but intellectually demanding sentences.

What this page owns—and what it does not

This page owns writing as a mechanism for learning subject content.

It does not replace writing instruction. Students still need explicit teaching in sentence construction, paragraph organisation, genre, grammar and revision. It does not replace the separate eduKateSG owners for self-explanation, retrieval practice, note-taking, learning journals or writing exemplars. How Writing Exemplars Work owns how models reveal craft choices. How Sentence Combining Works owns syntactic control through sentence combining.

The canonical question here is narrower: How can the act of writing about mathematics, science, history, geography or another body of knowledge make the learner understand that knowledge better?

Stage 1: Writing turns “I know it” into something inspectable

Before writing, understanding can feel global. “I get photosynthesis.” “I know why the war started.” “I understand fractions.”

Those feelings are difficult to test.

Ask for an explanation and the knowledge becomes granular. Which parts can be stated? Which relation is missing? Which vocabulary term is vague? Where does the sequence break?

Suppose a student says, “Increasing temperature makes particles move faster, so diffusion happens faster.” That sentence may reveal a reasonably coherent mechanism. Another student writes, “Temperature makes diffusion faster because heat speeds it up.” The second answer sounds plausible but leaves the causal model thin.

Writing creates an artefact that can be examined.

This is valuable for the learner too. The page becomes a mirror of the current knowledge structure.

Stage 2: Retrieval matters more than copying

Copying can produce neat notes without requiring the learner to reconstruct knowledge.

Writing to learn is stronger when at least some of the information must be retrieved rather than transcribed.

A simple sequence:

1. Read or study a short section.
2. Close the source.
3. Write the main mechanism from memory.
4. Reopen the source.
5. Mark missing or inaccurate parts.
6. Rewrite the explanation.

Now writing is doing two jobs: retrieval and reconstruction.

The learner experiences the difference between recognition—“yes, this looks familiar”—and recall—“can I produce the idea without seeing it?”

This is one reason a five-minute closed-book explanation may reveal more than twenty minutes of beautiful copied notes.

Stage 3: Selection forces the learner to decide what matters

A chapter contains far more information than can fit into a short explanation. The writer must select.

Selection is intellectually valuable when the prompt is well designed.

“Summarise the chapter” may produce arbitrary compression. “Explain the three steps without which the mechanism would fail” requires the learner to decide what is causally essential. “Choose the strongest evidence for the claim and explain why it is stronger than the alternative” requires evaluative selection.

Writing becomes a way of ranking information by relevance.

This is different from highlighting, where learners can mark too much and postpone the decision about importance.

Stage 4: Organisation changes what can be understood

Ideas do not live as isolated facts. Their relationships matter.

When writing an explanation, the learner must choose an order. Cause before effect. General principle before example. Claim before evidence. Observation before inference. Problem before solution.

A poor sequence can reveal poor understanding.

For example, a student knows four correct facts about the water cycle but cannot organise them into a process. Writing a causal sequence exposes that the problem is not missing vocabulary; it is missing relational structure.

The act of organising can itself improve understanding because the learner has to build a coherent map rather than retain a bag of statements.

Stage 5: Writing makes causal gaps visible

One of the most valuable words in writing to learn is because.

“The metal spoon became hot because heat travelled through it.”

That is still incomplete. How did the energy transfer? What properties of the material matter?

“Plants grow toward light because they need light.”

Again, a purpose-like statement may be substituting for mechanism.

A good writing prompt asks for the missing bridge:

cause → mechanism → observable effect.

When the learner cannot write the bridge, the gap is now visible enough to teach.

Stage 6: Comparison writing forces discrimination

Students often think they know two concepts because they recognise both names. Ask them to compare and the boundary becomes harder.

“Mass and weight are different” is a remembered statement. “Mass is the amount of matter and remains constant across locations; weight is a force that depends on gravitational field strength” requires relational discrimination.

Useful comparison prompts include:

• What is the same?
• What is different?
• Which difference matters most for solving this problem?
• Under what condition would the two behave differently?
• Why are they commonly confused?

Writing the comparison forces the learner to define concepts against each other, often producing more precise knowledge than studying each separately.

Stage 7: Prediction writing reveals the learner’s model before the answer arrives

Before a demonstration, worked solution or experiment, ask students to write what they expect and why.

The prediction itself is less important than the reason.

If the outcome differs, the learner can compare the original model with the evidence. Without the written prediction, the mind is good at saying, “I knew that.” With the prediction visible, the discrepancy becomes harder to erase.

A useful sequence is:

Predict → justify → observe → explain discrepancy → revise.

This uses writing to preserve the learner’s pre-instruction model long enough for conceptual change to become explicit.

Stage 8: Argument writing forces evidence-to-claim relations

Students can collect evidence without knowing how it supports a claim.

A short argument prompt asks them to connect:

claim → evidence → warrant.

In science: “Which model better explains the data?”

In history: “Which factor mattered most, and why?”

In geography: “Does the pattern support the proposed explanation?”

In mathematics: “Why is this method valid for every case under the stated condition?”

The writing reveals whether evidence is merely attached or actually used.

Stage 9: Writing in mathematics is not “English added to maths”

A useful mathematical writing task is tightly connected to mathematical reasoning.

Examples:

• explain why two methods give the same result;
• write a counterexample to a false rule;
• justify one transformation in a proof;
• describe what changes and what remains invariant;
• explain why a common wrong answer is tempting;
• state when a method would fail.

The goal is not elegant prose. It is explicit mathematical relation.

A student who can calculate but cannot state why a step is valid may have procedural knowledge without enough conceptual control.

Writing gives the teacher a way to see that distinction.

Stage 10: Writing in science can separate observation from explanation

Science learners often collapse observation and inference.

A writing frame can temporarily separate them:

We observed…

This suggests…

The model explains this because…

If the model is correct, we would also predict…

This makes epistemic roles visible. The observation is not the model. The model is not the evidence. The prediction is not yet an observation.

As students gain control, the frame can fade.

Stage 11: Writing in history can stop sources becoming quote containers

A student may insert a quotation and believe the evidential work is complete.

A writing-to-learn prompt can ask:

“What can this source support about the question, and what can it not establish?”

The learner now has to reason about provenance, context and claim fit.

Another useful prompt is to write the same conclusion twice: once using Source A and once using Source B. If the conclusion must change, the student begins to see that evidence shapes the claim rather than decorating it.

Stage 12: Short writing can outperform long writing when the thinking target is precise

Length is not automatically depth.

A 500-word generic response can contain more filler than a 70-word causal explanation. Writing-to-learn tasks should be long enough for the intended cognitive work and no longer by default.

Useful short forms include:

• one-minute explanation;
• three-sentence comparison;
• “because–but–so” relation;
• error explanation;
• prediction with justification;
• claim-evidence-warrant paragraph;
• exit ticket: “What changed in your understanding today?”

The teacher can use short writing frequently without turning every lesson into an essay lesson.

Stage 13: But “write anything” is not the mechanism

Free writing can support reflection or idea generation, but content learning is more reliable when the prompt directs the learner toward useful processing.

Compare:

“Write everything you know about ecosystems.”

with:

“Explain why removing one predator can change populations several links away in the food web.”

The second prompt selects causal structure.

Or:

“Write about fractions.”

versus:

“Explain why multiplying by a number between 0 and 1 can make a positive quantity smaller.”

The second prompt targets a misconception.

Prompt design is therefore not cosmetic. It defines the thinking job.

Stage 14: Audience can change the quality of explanation

Students often write differently when they imagine a real reader.

“Explain it to yourself” can produce compressed language because the writer already knows what is meant. “Explain it to a younger student who knows X but not Y” forces hidden assumptions into the open.

This does not mean every task needs an elaborate authentic audience. A simple imagined audience can be enough to increase explanatory completeness.

Useful audiences include:

• a younger learner;
• a classmate who missed the lesson;
• a sceptic who disagrees;
• a future version of yourself revising for an exam;
• a reader who can see the data but not the conclusion.

The audience changes what must be made explicit.

Stage 15: Writing can slow thinking productively

Fast classroom talk privileges quick retrieval and verbal confidence. Writing gives more time to hold, revise and inspect an idea.

A quiet learner may produce better reasoning in writing than in rapid discussion. A confident speaker may discover that fluent talk was hiding a gap.

That makes writing useful before discussion:

think → write → discuss → revise.

The written first step broadens participation and gives every learner something to bring into the conversation.

Stage 16: Writing can also overload working memory

The mechanism has limits.

A learner may be asked simultaneously to:

• remember new subject content;
• organise an argument;
• spell unfamiliar vocabulary;
• construct sentences in a second language;
• manage handwriting;
• remember paragraph conventions.

The result can be a writing bottleneck that obscures subject understanding.

Support should match the learning goal. If the goal is science reasoning, the teacher might allow bullet points first, provide essential vocabulary, accept a diagram plus explanation or separate concept feedback from language feedback. Later, language quality can be improved deliberately.

Reducing irrelevant load is not lowering the intellectual demand.

Stage 17: Feedback determines whether writing becomes another dead end

Students can write, hand in the page, receive a mark and never look at the reasoning again. The learning opportunity is then limited.

Feedback should change the next version.

Useful feedback points to the intellectual gap:

• “You named the cause; add the mechanism.”
• “The evidence is relevant; explain why it supports the claim.”
• “You described both processes; now state the critical difference.”
• “This example works; explain whether it proves the general rule.”

Then the student revises.

Writing becomes more powerful when it participates in a loop rather than a one-way submission.

Stage 18: Self-correction is part of the learning mechanism

Before teacher feedback, ask students to compare the response with a criterion or source.

For example:

1. Write the explanation from memory.
2. Reopen notes.
3. Underline accurate elements.
4. Circle omissions or misconceptions.
5. Add one correction in a different colour.
6. Rewrite from memory later.

The learner sees error as information.

The second attempt is crucial. If the corrected source remains visible, the learner may only copy the fix. Delayed rewriting tests whether the structure has changed.

Stage 19: Writing and self-explanation are neighbours, not the same owner

Some writing-to-learn tasks work partly because they prompt explanation. eduKateSG already has a separate self-explanation owner for the mechanism by which learners explain steps and relations to themselves.

Writing to learn is broader. It can include retrieval summaries, comparison, prediction, argument, source evaluation, error analysis, question generation and knowledge synthesis.

The overlap is real, but the canonical job differs:

• self-explanation: explain why or how the material works;
• writing to learn: use writing as an external cognitive workspace for several kinds of learning operations.

Clear boundaries prevent one article from swallowing an entire field.

Stage 20: Writing from memory is different from note-taking

Notes are often taken while information is present. Writing to learn can deliberately remove the source.

After a ten-minute lesson, ask:

“Without looking back, write the four ideas you think are essential and connect them with arrows or sentences.”

The learner now has to retrieve and structure.

Then compare with the original material. The mismatch is diagnostic.

This is especially useful for revision because students often overestimate what familiar notes mean they can reproduce independently.

Stage 21: The Clementi-floor example—why the page changes the lesson

A Secondary 3 student, Iona, is revising electrolysis. She has read the textbook twice and highlighted half the page. It feels familiar.

Her teacher asks her to write, without notes: “Why does copper form at one electrode in this setup? Use ions, charge and electron transfer.”

Iona writes the first line easily, then stalls at electron transfer. She realises she has memorised electrode names without understanding the reduction step.

She checks the text, corrects the model, closes the book and writes again.

Ten minutes later, she has written fewer words than her original notes contained. But the useful learning event was stronger because the writing forced the missing relation into view.

Stage 22: Another example—mathematics error explanation

A class is shown the false statement:

(a + b)^2 = a^2 + b^2

Instead of only correcting it, students write: “Why might someone believe this? Show the missing mathematical term and explain where it comes from.”

The task turns an error into conceptual analysis. Students must expand, identify 2ab, and explain why distribution creates the cross terms.

That is writing to learn because the writing forces structure, not because a paragraph is inherently educational.

Stage 23: Another example—history claim revision

Students answer: “The policy failed because it was unpopular.”

The teacher provides two sources and asks them to rewrite the claim twice:

• once using only Source A;
• once after considering Source B.

If Source B complicates the picture, the revised sentence may become: “Public opposition contributed to the policy’s failure, but administrative resistance and implementation problems also limited its effect.”

The act of rewriting is a knowledge revision, not merely a stylistic revision.

Stage 24: What the meta-analysis suggests

A major 2020 meta-analysis by Steve Graham, Sharlene Kiuhara and Meade MacKay examined 56 experiments in Grades 1–12 where students wrote about content in science, social studies or mathematics. On average, writing about content improved learning compared with control conditions, with an overall effect size around 0.30.

That is encouraging, but the average should not be turned into a promise that every writing activity works. Effects varied substantially across studies, and some individual results were negative. The meta-analysis also could not cleanly identify one task feature that explained all the variability.

The sensible interpretation is:

Writing about content can improve learning on average, across several subjects and age groups, but task design and context still matter, and “writing” is too broad a label to guarantee benefit.

That is exactly why mechanism matters more than slogans.

Stage 25: Current research signals show the question is still active

The Institute of Education Sciences is currently supporting work related to writing and learning. Its active meta-analysis of college writing interventions explicitly includes course learning outcomes and writing-to-learn approaches. Another IES project, College Ready: Reading and Writing to Learn, runs through December 2026 and is developing and assessing an intervention for high school students with or at risk for disabilities.

These projects are not themselves proof that a particular classroom routine works. Development studies and ongoing meta-analyses should not be reported as completed effectiveness evidence.

They are useful as a current signal that writing is being studied not only as a composition outcome but as a route into learning, source use and knowledge construction.

Stage 26: Failure mode—copying masquerades as writing to learn

Students copy slides, rewrite the textbook or transcribe a model answer. The page fills. The hand gets tired. Learning may barely change.

Copying can have a limited role when accuracy of notation or language matters, but it should not be confused with generative writing.

A quick repair is to remove the source for part of the task. Ask the learner to reconstruct first, then check.

If the learner never has to decide what to say, writing is doing little cognitive work.

Stage 27: Failure mode—quantity becomes the target

“Write one page” is easy to assign and easy to measure. It can produce filler.

A stronger target names the thinking:

• explain two causal links;
• compare three features;
• justify one method;
• identify one limitation;
• revise one misconception;
• connect evidence to claim.

Length can follow from the job.

Stage 28: Failure mode—every lesson ends with a generic reflection

“What did you learn today?” can be useful occasionally, but repeated generic reflections often produce equally generic answers: “I learned about cells.”

Make the prompt diagnostic:

• “What idea from today would be easiest to misunderstand, and why?”
• “Which step in the process causes the next step?”
• “What evidence changed your prediction?”
• “Write one rule and one case where it does not apply.”

The specificity makes writing cognitively useful.

Stage 29: Failure mode—grading every piece makes students write defensively

If every short learning write is heavily graded for grammar, spelling, structure and content, students may focus on surface correctness and risk avoidance.

Some writing-to-learn should be low stakes. The teacher can sample reasoning without marking every sentence. Students can self-check. Peers can compare explanations. A few pieces can be developed into polished work later.

The distinction between writing to think and writing to publish matters.

Both are valuable. They need different feedback intensity.

Stage 30: Failure mode—writing replaces necessary practice

A learner cannot become fluent in algebra by writing essays about algebra instead of solving problems. A student cannot learn pronunciation only by writing about speech. A laboratory skill cannot be acquired through prose alone.

Writing is one learning operation among others.

Use it when explanation, organisation, comparison, argument or reflection on knowledge is part of the target. Combine it with direct instruction, worked examples, deliberate practice, retrieval, discussion and feedback as needed.

The question is not “How do we add writing?” It is “Where would writing expose or strengthen the reasoning this lesson needs?”

Stage 31: Failure mode—language accuracy obscures conceptual diagnosis

A multilingual learner writes an awkward but scientifically accurate explanation. If the teacher marks only grammar, the subject reasoning disappears.

Separate feedback dimensions when useful:

Science: Is the mechanism correct?

Language: Is the explanation precise and comprehensible?

Both matter. They can be improved in sequence rather than collapsed into one judgement.

This is especially important when the purpose of the writing is to learn the subject rather than to assess polished language production.

Stage 32: Failure mode—students never see what good reasoning looks like

If a prompt asks for explanation but the student has only seen factual answers, the task can feel mysterious.

Use contrasting examples. Show two short responses containing similar vocabulary. Ask which better explains and why. Highlight the causal bridge, evidence relation or comparison structure.

This is where writing exemplars can support the mechanism without becoming the same owner. Exemplars reveal what the target looks like; writing to learn asks the learner to use writing to reconstruct knowledge.

Stage 33: Failure mode—revision becomes proofreading only

Students often hear “revise” and check spelling.

For writing to learn, revision should first ask whether the idea changed:

• Is the explanation more causally complete?
• Is the comparison more discriminating?
• Is the evidence better connected?
• Is the claim more accurately bounded?
• Has a misconception been removed?

Proofreading matters later. Conceptual revision comes first when the writing’s purpose is learning.

Stage 34: The teacher route—choose the thinking verb before the writing format

Plan backwards:

What should students do mentally? Retrieve, explain, compare, predict, evaluate, connect, justify, revise?

Then choose the shortest writing form that makes that operation visible.

Examples:

• retrieve → 90-second closed-book summary;
• explain → mechanism paragraph;
• compare → two-column contrast plus synthesis sentence;
• predict → prediction and reason before demonstration;
• evaluate → claim-evidence-limit paragraph;
• justify → one-step proof explanation;
• revise → before/after explanation with change note.

This keeps writing subordinate to learning rather than making writing quantity the objective.

Stage 35: The learner route—WRITE, CHECK, REWRITE

A learner can use a three-part cycle:

WRITE: Without looking, explain or solve the prompt in your own words.

CHECK: Compare with notes, textbook, worked solution or teacher feedback. Mark missing relations, not just missing words.

REWRITE: Close the source and produce the improved version again.

The final step matters. It tests whether the correction has entered memory rather than remaining visible on the page.

Stage 36: The parent route—ask for a tiny explanation, not another hour of copying

Parents do not need to mark essays.

After revision, ask the child to write three to five sentences answering a precise question:

• “Why does this happen?”
• “What is the difference between these two ideas?”
• “What evidence supports this claim?”
• “Why is this step valid?”

Then ask the child to compare the answer with the source and identify one improvement.

This is often more diagnostic than asking whether the notes are complete.

Stage 37: The ordinary-weekday transfer test

Writing to learn has transferred when the learner begins to use it voluntarily as a thinking tool.

On a normal weekday, does the student write a quick explanation to test whether a concept is understood? Sketch the causal chain before an essay? Write a prediction before checking the answer? Record why a mathematics error occurred rather than merely correcting it?

The mature form is not “teacher assigns more writing.” It is learner reaches for writing because writing helps think.

A compact classroom sequence

For one concept:

1. Teach or study.

2. Close the source.

3. Prompt one high-value operation. Explain, compare, justify or predict.

4. Write briefly. Enough to expose the structure.

5. Check against evidence or a model.

6. Revise the idea. Not only the wording.

7. Return later with a new prompt. Test retention and transfer.

This sequence can take ten minutes and does not require a formal essay.

Evidence caveats that matter

The positive average in writing-to-learn research does not tell us that every learner needs the same prompt, length or feedback. Studies vary. Writing imposes transcription and language demands. Subject tasks differ. Some learners may benefit from oral rehearsal or diagrams before prose. Effects can depend on prior knowledge and implementation.

The strongest claim is therefore conditional: writing can support learning when it causes learners to engage in useful processing of the content and when its language demands do not overwhelm the intended cognitive job.

That is more precise—and more useful—than “writing is good for learning.”

FAQs

Is writing to learn the same as teaching writing?

No. Teaching writing aims to improve writing knowledge and skill. Writing to learn uses writing as a tool for learning another body of knowledge. A classroom can do both, but the immediate goals differ.

Does copying notes count?

Usually not in the strong sense used here. Copying may support accuracy or provide a reference, but it requires less retrieval, selection and organisation than writing from memory or responding to a reasoning prompt.

How long should a writing-to-learn task be?

As long as needed for the thinking operation. Many useful tasks can be completed in two to five minutes. Longer writing can be valuable when synthesis or argument requires it.

Should grammar be corrected?

If language development is also a goal, yes—but consider sequencing feedback. When the purpose is subject learning, diagnose the conceptual reasoning first so grammar does not obscure whether the student understood the idea.

Does the evidence show a large effect?

A 2020 meta-analysis of 56 experiments found an average effect size of about 0.30 across science, social studies and mathematics in Grades 1–12. That is a positive average, not a guarantee for every task or setting, and effects varied across studies.

What kind of prompt is best?

There is no single universal best prompt. Strong prompts direct attention to the subject’s learning goal: explain a mechanism, compare confusable concepts, justify a method, predict an outcome, connect evidence to a claim, or diagnose an error.

Can students type instead of handwriting?

Yes, depending on the goal and learner. The mechanism here is generative production and organisation of ideas, not a claim that handwriting is always superior. Choose the mode that lets the learner perform the intended thinking without unnecessary barriers.

Can AI help with writing to learn?

It can provide feedback or questions, but the learner must still perform the cognitive operation that is meant to be learned. If AI produces the explanation before the student attempts it, the writing may cease to function as retrieval and construction. A safer sequence is attempt first, then use tools for critique or checking.

Is a learning journal writing to learn?

It can be, if entries require useful processing. A journal that merely records “what we did today” may be less powerful than one that asks the learner to explain a change in understanding, analyse an error or connect ideas.

What is the quickest way to try this tomorrow?

At the end of one lesson, close all notes and ask one precise question that requires explanation rather than recall. Give students three minutes to write. Then let them check the source and revise one sentence. Read a few to decide what needs reteaching.

Sources and further reading

• Graham, Kiuhara & MacKay, The Effects of Writing on Learning in Science, Social Studies, and Mathematics: A Meta-Analysis, Review of Educational Research (2020): https://doi.org/10.3102/0034654320914744
• Institute of Education Sciences, Meta-analysis of College Writing Interventions on Writing and Learning Outcomes: https://ies.ed.gov/use-work/awards/meta-analysis-college-writing-interventions-writing-and-learning-outcomes
• Institute of Education Sciences, College Ready: Reading and Writing to Learn: https://ies.ed.gov/use-work/awards/college-ready-reading-and-writing-learn
• eduKateSG, How Writing Exemplars Work: https://edukatesg.com/2026/09/16/how-writing-exemplars-work-why-reading-like-a-writer-makes-hidden-craft-choices-visible/
• eduKateSG, How Sentence Combining Works: https://edukatesg.com/2026/09/15/how-sentence-combining-works-why-joining-short-sentences-builds-syntactic-control-and-writing-quality/

Where this sits in the eduKateSG map

This article belongs under the existing How X Works Hub. It owns writing to learn as a mechanism: using writing to retrieve, select, organise, explain, compare, justify and revise knowledge.

The shortest useful version is this: writing helps learning when the learner has to use the page to do intellectual work—not merely fill the page.