Harvard Professor Explains The Rules of Writing — Steven Pinker

David Perell

Abstract

Forty-three minutes on why bad writing happens, with an answer that removes the usual villain: almost none of it is deliberate obscurity. It comes from capable people who cannot subtract a fact from their own brain — the curse of knowledge — which is named here as the single biggest flaw in writing and communication.

• The curse cannot be detected from inside the head that has it, and empathy helps without being sufficient.
• The only dependable fix is handing the draft to an intelligent reader who is not a peer in your field.
• Understanding is not made of words; language conveys ideas that are usually sensory, so prose has to be concrete.
• A generalisation without an example and an example without a generalisation are both close to useless.
• Every extra word is extra cognitive processing for the reader, which is why forcing a fixed length works.
• Earlier writers were vivid partly because no cliche was available to reach for.

Everything below follows the conversation in order, from the diagnosis through the visual mind and the speech-writing gap, into rhythm, brevity and humour, out through why non-writers wrote so well, informalisation, and finally machine prose. For anyone building a personal brand the operative instruction is the least comfortable one: you cannot check your own work for the thing most likely to be wrong with it.

Chapter summaries

00:38 - Bad writing is rarely deliberate obscurity

Bad writing is rarely a deliberate choice to obscure; most of it comes from capable people who simply cannot express themselves clearly.

"When I uh pose the question why is there so much bad writing? Why? Why"

— 00:38

Never attribute to malice what stupidity explains — but it is a specific stupidity: not knowing where the audience is coming from. Brilliant people with plenty to say and no desire to obfuscate still write badly, which rules out the usual explanation. The people meant are not stupid people; it is a narrow blindness.

03:01 - The curse of knowledge is not knowing what it is like not to know

The curse of knowledge is the difficulty of knowing what it is like not to know something you know — you cannot subtract a fact from your own brain.

"curse of knowledge term from economics is um the difficulty that we all have in"

— 03:01

You cannot subtract a fact from your own brain, which is why the effort of imagining the reader fails in a specific and repeatable way. The mechanism is subtraction, and it is not available.

03:30 - The curse surfaces as acronyms, jargon and hidden abstractions

The curse surfaces as unexplained acronyms, clique jargon, and abstractions that hide a concrete thing.

"abbreviations and acronyms that no one has any way of knowing. You have jargon that"

— 03:30

An abstraction compresses an enormous body of knowledge, so the writer forgets it needs unpacking for anyone who has not done the compressing. The level of the stimulus was proportional to the intensity of the reaction means children look longer at a rabbit than at a truck.

04:09 - The curse cannot be detected from inside the head that has it

The curse of knowledge — egocentrism, an absent theory of other minds — is the single biggest flaw in writing and communication.

"Egoentrism, absence of a theory of mind. That is a theory of what's going on"

— 04:09

It is named as the one flaw he would pick above all others if only one could be named — offered as a personal identification rather than a measured ranking. The thing that makes it the biggest is that it is invisible from inside the head that has it.

05:11 - The only dependable fix is a real reader outside the field

The only dependable fix is to hand the draft to a real reader who is intelligent but not a peer in your field.

"mother was alive, I would always show her a uh a draft of my book,"

— 05:11

Hand the draft to somebody intelligent who is not a peer in your field. Even fellow experts are insular — step outside a small circle and the same prose stops being intelligible.

07:29 - The mind runs on vision, so good writing is concrete

Because the mind runs heavily on vision, writing well means being concrete and letting the reader see what you mean.

"a good way of describing it? Yes. And that would be my the probably the"

— 07:29

Understanding is not made of words; language is a means of conveying ideas that are usually sensory. The words are the delivery mechanism rather than the cargo, which is why an abstract sentence can be perfectly grammatical and still arrive empty.

08:31 - The reader has to be able to form a mental image

Let the reader form a mental image; do not write "stimulus" for a rabbit, or "paradigm" where no image is possible.

"constantly allowing your reader to be able to form a mental image uh based on"

— 08:31

Not stimulus where the thing is a rabbit, and not a paradigm where no image is possible. Worked examples and visual metaphor do the work — and earlier prose reads as vivid partly because it was written before abstraction was available.

10:48 - Abstractions are essential inside a profession

Abstractions are essential inside a profession; the danger is only at the moment of conveying them outward.

"right. So they they do make it in fact they're essential for doing the work"

— 10:48

The abstraction is not the problem; the audience boundary is. Inside the profession the term is a shorthand everyone shares; outside it, the same term is a wall with no door in it.

11:49 - Speaking comes naturally and writing does not, for structural reasons

Speaking comes naturally and writing does not, and the reason is structural, not a lack of talent.

"anyone else. Hm. What do we need to know about writers about how hard writing"

— 11:48

In speech, shared context lets vague words work. In writing the reader is wrenched from every context and everything has to be on the page — which makes the difficulty structural rather than a matter of talent.

13:39 - Speech gives real-time feedback that writing cannot

Speech gives real-time feedback that writing cannot; the furrowed brow tells you the curse of knowledge has struck.

"when you're speaking, you know that you're speaking to someone. You know their idiosyncrasies. You're"

— 13:39

A furrowed brow tells you the curse has struck, and writing offers nothing equivalent. The absence of that feedback is what makes the outside reader necessary rather than helpful.

14:19 - A generalisation without an example is close to useless

A generalisation without an example and an example without a generalisation are both close to useless.

"one? I think generalizations without examples and examples without generalizations are both useless. Yes. Well,"

— 14:19

The same holds in reverse, for an example with no generalisation attached to it. They are two halves of a single unit rather than two options a writer picks between.

15:36 - Writing lives in the trade-off between context and compression

Writing lives in the trade-off between context and compression — examples supply context, generalisations compress.

"my favorite ways to think about this is that there's always a trade-off between context"

— 15:36

Examples supply context and generalisations compress. Every sentence sits somewhere on that trade-off, which makes it a design decision rather than a stylistic one. Choosing wrong in either direction produces prose that is either bloated or inert.

16:11 - A claim is only understood once an example is attached

A claim only becomes understood once an example is attached; before that the reader can nod without grasping it.

"example, if I were to say that familiar words don't have to refer to the"

— 16:11

The nodding is the dangerous part, because it looks like comprehension from both sides. The writer reads agreement and moves on, and the reader discovers later that nothing was retained.

20:07 - Prose has a sound, and reading aloud is the test

Prose has a sound; read it aloud, and if you cannot say it smoothly the reader cannot mentally sound it out smoothly either.

"ago. When you're writing, like if I were to basically map out almost like a"

— 20:06

The inner ear runs even over silent text, so an unspeakable sentence trips a reader who never opens their mouth. Reading or mumbling one's own drafts aloud is presented as standard advice he did not invent.

21:20 - Rhythm and sound get attended to deliberately

Attend to rhythm and sound — disrupt the beat and you interfere with reading; thin out sibilants; spend alliteration sparingly.

"Um the aesthetics sometimes come from even things like um uh paying attention to the"

— 21:20

Disrupt the beat and you interfere with reading. Sibilants get thinned out and alliteration gets spent sparingly, which are unusually specific instructions for something usually left to instinct.

22:36 - Bad academic prose is a triple waste

Bad academic prose is a triple waste: brilliant work locked in a closed circle, effort spent decoding it, and beauty foregone.

"Why do you think that you as much as anybody have been the person to"

— 22:36

Publicly funded knowledge that is inaccessible fails its own purpose, and prose read six times over may still be misread at the end of it. Rereading the same paragraph five or six times to review a paper, and possibly still getting it wrong, is the concrete cost.

25:14 - Children explain freshly because they lack the abstractions

Children's explanations are fresh because, lacking abstractions, they must appeal to something that can be seen.

"Yeah. What is it that you love so much about children's explanations? Things like clouds"

— 25:14

Lacking the abstractions, they have to appeal to something that can be seen. Children cannot think inside the box because they do not know the box exists, and their phrasing has been mined as entertainment for generations.

27:47 - Humour depends on freshness and dies when strained

Humour depends on freshness and dies when strained or overused; it has much in common with good writing.

"to to much hilarity. How does humor factor into all this into writing? Yeah, exactly."

— 27:47

The overlap with good writing is more than coincidental — both are destroyed by the same thing. Strain shows in a joke a beat before it shows in a sentence, which makes humour the earlier warning of the two.

29:44 - Forcing prose into a fixed length improves it

Forcing prose into a fixed length improves it as if by magic.

"Now again it's a beautiful example of itself because there are no needless words in"

— 29:44

Brevity is the soul of wit is an example of itself, and omit needless words is the same lesson restated and also an example of itself. Forcing prose into a fixed length improves it as if by magic.

30:46 - Brevity works because every extra word costs the reader

Brevity works because every extra word is extra cognitive processing for the reader, and it forces concrete language.

"But what I find is often just the exercise of squeezing it into that uh"

— 30:46

Fewer syllables means less work to receive the same message, and the constraint pushes out woolly idiom and cliché on its way through. Squeezing an article down and finding the meaning survives — tighter than before — is the felt version of the argument.

32:36 - Some of the best prose came from people who were not writers

Some of the best-written prose in history came from people who were not professional writers at all.

"So I went to uh Rockefeller Center the other day and there's this called like"

— 32:36

An inventor, nineteenth-century thinkers, figures known for entirely other work: none of them wrote for a living, and all of them could write like angels. The trade is not the qualification.

34:54 - Earlier writers had no cliche available to reach for

Exploring new ideas with no cliche available forced earlier writers into vivid images and metaphor.

"the the third reason is one that came up earlier in our conversation. Namely, since"

— 34:54

They trained on the classics, had no other way to present themselves, and explored new ideas with no cliche available — which forced them into vivid images and metaphor. The absence of ready-made phrasing was doing the work.

35:35 - A century of informalisation has reshaped how prose is judged

A century-long process of informalisation has reshaped how prose is judged.

"that is that there's been a um a process that's been going on for longer"

— 35:35

A century-long process of informalisation, flowing from democratisation and a romantic valuing of authenticity and spontaneity over putting on airs. It is a change in the judging standard rather than in the writing.

37:34 - Fancy language now reads as pompous

Fancy language now reads as pompous, and the conversational vernacular has come to define modern prose.

"I think many of us in having if we had to put the effort into"

— 37:34

Which register counts as good writing is a historical fact rather than a fixed one, and the current one arrived within living memory. What reads as sincere now read as careless a century ago, and nothing about the sentences changed.

38:36 - Machine prose is sound in structure and generic in effect

A large language model's prose is peculiar: sound and plain in structure, yet so generic you can almost recognise it.

"the output of LLM is is peculiar in one sense. It's well written in the"

— 38:36

Sound and plain in structure, yet so generic it is almost recognisable as such. It could perhaps be prompted toward freshness, and it is designed as a pastiche of billions of examples — and it may read as sound because human feedback hammered it into shape, unlike unedited expert prose.

40:20 - Averaging may clarify prose the way it flatters a face

Speculatively, averaging may clarify prose the way it flatters a face — a composite is cleaner, though not beautiful.

"uh and this is completely speculative but we know in visual beauty often a composite"

— 40:20

A composite face is cleaner without being beautiful, and the analogy is offered explicitly as speculation — which is the right register for a claim nobody has tested.

41:15 - A whole career of books would now weight pattern extraction far more

Rewriting a career of books today, he would give far greater weight to pattern-extraction from massive input.

"That's a hypothesis. You've written nine books. You're working on your 10th now. with LLMs"

— 41:15

It is a working scientist revising his own position in public, which is rarer than it should be. The revision is specific rather than gestural: a named weight on a named factor, which is the form a real change of mind takes.

42:33 - A human mind is not a language model

A human mind is not a large language model, because a child learns from a situated world, not from oceans of text.

"Mhm. Now, I don't think that shows that the human mind is a large language"

— 42:33

A child learns from a situated world rather than from oceans of text, so the two are not the same kind of thing however similar the outputs look. The closing distinction is about the input rather than the architecture.

Personal Branding Lessons

A conversation about a flaw nobody can see in themselves. The moves below are the working corrections.

Hand the draft to an intelligent reader outside your field

The curse of knowledge cannot be detected from inside the head that has it. Even fellow experts are insular enough to be useless as a test — step outside a small circle and the same prose stops being intelligible. Somebody smart and unfamiliar is the only reliable instrument. 05:11

Let the reader see what you mean

The mind runs heavily on vision, and understanding is not made of words. Write the rabbit rather than the stimulus. Where no image is possible, a worked example or a visual metaphor does the job that the abstraction was quietly failing to do. 08:31

Attach an example to every generalisation

A generalisation without an example and an example without a generalisation are both close to useless — two halves of one unit rather than two options. Before the example arrives, a reader can nod along without grasping anything, and the nodding looks like comprehension from both sides. 14:19

Read it aloud and listen for the places you stumble

Prose has a sound. If you cannot say a sentence smoothly, the reader cannot mentally sound it out smoothly either. The mouth turns out to be a better instrument for finding a bad sentence than the eye, and it costs nothing to use. 20:07

Thin out the sibilants and spend alliteration sparingly

Disrupt the beat and you interfere with reading itself. These are unusually specific instructions for something normally left to instinct, which is what makes them usable — they can be applied to a draft rather than hoped for during one. 21:20

Force the prose into a fixed length

Squeezing a piece into a set length improves it as if by magic. The constraint does the editing that judgement was avoiding, and it works regardless of whether the writer can see what needed to go. 29:44

Omit needless words, because every one costs the reader

Every extra word is extra cognitive processing, which makes brevity a courtesy with a measurable basis rather than a style preference. It also forces concrete language, because vagueness is what takes up the room. 30:46

Explain it the way a child would, with something visible

Children explain freshly because, lacking the abstractions, they have to appeal to something that can be seen. They cannot think inside the box because they do not know the box exists — which is the state the expert is trying to get back to. 25:14

Questions

Each answer ends at the moment in the recording where it is given.

Why is so much expert writing so hard to read even when the writer is brilliant?

Because bad writing is rarely a pose. It usually comes from capable people with plenty to say who simply do not know where their audience is coming from — a narrow blindness, not general incapacity. The expert forgets that the acronyms, jargon and abstractions that are second nature inside the field mean nothing outside it, and never notices the gap because it is invisible from the inside. 00:38

What is the single biggest flaw in writing?

The curse of knowledge — the near-impossibility of imagining what it is like not to know something you already know, so you cannot subtract a fact from your own brain. Offered as a personal identification rather than a measured ranking, it is named as the one fault above all others, and it goes by harsher names too: egocentrism, the absence of a theory of what is happening in other minds. Almost all bad writing traces back to it. 04:09

How do I tell whether my own writing is unclear when it feels obvious to me?

You largely cannot, from the inside — which is the whole trap. What seems so obvious that you never question whether others share it is exactly what turns out not to be shared, and cultivating empathy helps but does not suffice. The only dependable fix is to hand the draft to an intelligent, well-read reader who is not a peer in your field and watch where they stall. Mining what your audience does not yet know beats trusting your own sense of clarity. 04:41

Why does writing feel so much harder than talking?

Because conversation starts from common ground and writing starts cold. Two people talking already know why they are there and can lean on vague words the setting makes clear; a reader is wrenched from all context and must pick everything up from the page, perhaps in another country, perhaps after the writer is dead. Speech also returns instant feedback — the furrowed brow that signals confusion — while a page shows nothing back at all. 11:49

Should I write in generalisations or in examples?

Both, together — a generalisation without an example and an example without a generalisation are close to useless apart. Abstract words are too imprecise to fix a single meaning, so the example is what pins the claim down; without it a reader can nod and agree while understanding nothing. Good writing is the movement between context, which examples supply, and compression, which generalisations provide. A repeatable structure helps you pair the two every time. 14:19

How do I make my sentences more pleasant to read?

Read them aloud, or at least mumble them, because the inner ear runs even over silent text — if you cannot say a sentence smoothly, a reader cannot sound it out smoothly either. Attend to rhythm, thin out runs of hissing sounds by swapping a synonym, and spend a little alliteration for a spark of pleasure. The one caution is restraint: make the ornament too conspicuous and it starts to feel forced. 20:07

Why does forcing myself to cut words improve the writing?

Because every extra word is extra cognitive processing for the reader, so a shorter path to the same message is simply less work to receive. A hard length limit acts as a forcing function: it strips the words you were sure were perfect, and the meaning survives tighter, while the constraint pushes you out of woolly idiom and cliché into concrete language. The improvement often feels like magic precisely because the cut words turn out to be expendable. Short-form discipline trains the same reflex. 29:44

Why does writing from a large language model sound clear but feel generic?

Because it is competent and averaged at once. The sentence structure is plain and sound, the progression of ideas orderly, with a tidy opening and closing — good writing in that narrow sense. But it is built as a pastiche of billions of examples, and averaging vast input yields the middle: clear, but so prosaic that you can almost recognise the source. The sameness is not a bug; it is what the design produces. 38:36

Can an AI assistant be trained to write with real style?

It is left as an open question. Prompting it away from plainness might help, but freshness and owned voice are not what the machine is designed for — it is a composite, and a composite is clear rather than beautiful. The most a hopeful analogy offers, and it is flagged as explicitly speculative, is that blending removes the worst constructions and leaves something clean, though not stylish. Real style still reads as the uncopyable work of one person. 39:10

Sources

Harvard Professor Explains The Rules of Writing — Steven Pinker

The sections above follow the episode's own order. Each timestamp is the point in the recording where the idea is discussed, so the post reads as a map of the conversation rather than a rearrangement of it.

Quotations come from the episode transcript with spoken filler ("okay", "right", "you know", "um", "actually", "kind of", "sort of") and false starts removed. No words are added: square brackets mark anything inserted for sense, and an ellipsis marks any cut. British spellings are restored, and speech-to-text errors were repaired against context only where the intended word is unambiguous. Where a line appears both in a cold-open teaser and in its place in the conversation, the timestamp cites the second — the point at which it is actually said.

Quotes are stamped with the moment they are said rather than with a speaker's name.