pg

Write Like Paul Graham

Use this when someone asks for an essay, article, or blog post "like Paul Graham" or "in PG's style." First settle the inline-format house style (below), write the piece in that voice, then revise out the AI tells using the checks further down.

First, pick an inline-format

PG's own prose uses almost no inline emphasis — but a blog post usually wants a few scan anchors, and how heavy to go is the reader's call, not yours. So before you write (or before a rewrite pass), ask the user with the AskUserQuestion tool which house style to apply. Offer these four, in this order, and attach a preview of the same short snippet rendered under each so they can compare:

1. Bold takeaway + italic terms (recommended default) — one bold punchline per section, the single line the eye should catch, plus _italic_ on a coined or key term's first mention only. About one bold per section. Stays mostly PG-plain.
2. Bold lead-ins (definition style) — open key sentences and list items with a bold label, so concepts scan down the left edge. Most skimmable, least PG; reads like docs.
3. Italics only (PG-light) — _italic_ for term-introduction and contrast, no bold at all. Closest to PG's own habit.
4. Code spans + structure only — no bold or italic. Scannability comes from ## headers, > blockquotes, and `identifier` code spans. Purest PG, lowest skim-ability.

If the user doesn't care, default to option 1. Whichever they pick, apply it consistently and keep it sparing: emphasis on every line is emphasis on none. Never bold a whole paragraph — bold at most one sentence, and make it the takeaway. Italicize a term only on its first use.

The voice

Sound like a smart friend explaining something at a bar, not a writer performing. Reason out loud in front of the reader: start from one plain observation, hit a snag, correct yourself, and land on a claim that feels like a thing you both just discovered rather than a thesis you walked in holding. Keep the prose short, flat, and Anglo-Saxon, so the rare crude or vivid word lands hard. Concede the obvious objection, then turn it over. Talk straight to "you." Hedge with "I think" and "my guess is" exactly where a weaker writer would bluff. And pin every abstraction to something physical — a leaking roof, 52 weekends, a hand-crank engine. The whole move is reducing a big mysterious thing to one mundane sentence.

Sentence mechanics

• Default to short, plain declaratives. Then vary length on purpose: build a long clause-stacked sentence, then chase it with a two-to-four word verdict. "Hard, but doable." "That's nonsense."
• Use fragments as punctuation for emphasis. A single word standing alone: "Why?" "I wish." "Then I had kids."
• Open sentences with conjunctions constantly — And, But, So, Plus, Or, Yet, Indeed, Which. This is load-bearing connective tissue, not a flourish.
• Lean on parataxis: set independent statements side by side instead of nesting clauses. The argument advances by addition.
• Reserve the em-dash for two jobs only: an abrupt mid-sentence aside, and a reversal that refines what you just said.
• Use parentheticals for a second thought or a deadpan correction. "(But she never does.)"
• Use the colon as a pivot from setup to payoff, often to deliver the thesis itself.
• Render lists as semicolon-chained vignettes or stacked imperatives, roughly equal in length, building toward a punchline.

Diction

• Default to plain, monosyllabic, Anglo-Saxon words. When you must use an abstraction, cash it out immediately in a concrete image or a named example.
• Allow exactly one sharp or crude word per stretch — "suck," "bullshit," "crap" — and let it land because everything around it is so plain. Crudeness is a precision tool, not a register.
• Ground everything in concrete nouns and brand names: Aeron chairs, rolly bags, 52 weekends, Baskin-Robbins, a 4K TRS-80. The props do the persuading.
• Drop a technical or scholarly term only when it's the precise word, and don't over-explain it. Gloss lightly if at all.
• Use contractions everywhere. It's spoken language on the page.
• Address the reader as "you," as a specific person facing the choice. Slide between "you," "one," and a raw first-person "I" confession.
• Coin your own term and then wield it as if it always existed ("the Blub paradox," "maker's schedule").
• Use precise-sounding numbers as anchors: "ahead of 95% of writers," "52 weekends," "67 minutes."
• Reach for the homely "X is the Y of Z" equation: "Informal language is the athletic clothing of ideas."

How an essay is built

An essay is something you write to figure something out, so build it as live reasoning, not a delivered conclusion. Don't open with the thesis you'll prove; open with a question or observation that genuinely puzzles you, and let the real thesis emerge late — often after you float a wrong guess and knock it down ("One possible explanation is... But this isn't true"). The shape is wander-then-converge: meander through facets, examples, and digressions, then pull the threads together in a final movement. You can even confess the wandering. Plant one governing analogy early (the hand-crank engine, sailing upwind) and return to it as the spine. For longer pieces, chunk under short bare section headers — often one word — and inside each run the same loop: state the principle, anchor it in a named real example, extract the lesson. Push qualifications and attributions into endnotes so the main line stays clean. Where you can, let the essay enact its thesis: a piece about brevity is brief; a piece about discovery visibly discovers.

Openings

• State a flat, deflationary thesis as bare fact, then immediately complicate it. "Life is short, as everyone knows. Is life actually short...?"
• Open on a concrete autobiographical scene or dated memory that postpones the argument and earns trust first. "When I finished grad school I went to art school to study painting."
• Ask the reader a direct, conspiratorial question that assumes shared complicity. "Remember the essays you had to write in high school?"
• Concede a cliché, then interrogate whether it's actually true.
• Narrate the moment of insight. "I finally realized today why..."
• Offer a "trick" or recipe and deliver the whole thesis before the second sentence.
• No throat-clearing. The hook is in the first sentence or two.

Endings

• Collapse the argument into a compressed triad of imperatives or a single rule of thumb, then loop back to the opening, now earned.
• Throw a rhetorical question back at the reader as a dare. "How hard is that?" "Why not by you?"
• Zoom out from the specifics to a wider historical or moral view, landing hopeful and forward-looking.
• Refuse the grand finish. Deflate into something modest, wry, or practical — a small ask, a deadpan tip.
• Resolve the governing analogy one last time so the metaphor closes the loop.
• Land a single sharp aphorism and stop, pushing any softening counter-case into a footnote.
• Swerve from sober candor to brisk encouragement and hand agency back. "If you want to do it, do it."

Rhetorical moves

• Concession-then-reversal — the workhorse. Grant the objection generously, then turn it. "On the whole his advice is good... But there is a contradiction."
• Voice the reader's objection in their own words, often as a quoted question, then answer it. "Really? How do you make them?"
• Pose a rhetorical question and answer it curtly. "How do you figure out what customers want? Watch them."
• Reframe the question itself from first principles — not "is this taking over the world" but "how big could it get if the founders did the right things."
• State a surprising claim flatly, as if obvious. "Prestige is just fossilized inspiration."
• Use the "X is just Y" / "X is really Y" reframe to deflate a mystified abstraction into something mundane.
• Use "not X, but Y" to sharpen a claim — surgically, to correct one misconception, never as filler.
• Hedge deliberately — "I think," "my guess is," "probably," "I suspect" — placed exactly where bluffing would be tempting.
• Use an extended analogy as the actual argument, not decoration; sometimes dismantle it later. "This metaphor doesn't stretch that far."
• Lower your own status for credibility. "We felt pretty lame at the time."
• Drop a deadpan aside that punctures the prose. "we, ah, skipped all that."

Never do this (the AI tells)

• Rule-of-three everywhere. Don't force every list and adjective into tidy triads. Let a list be as long as the idea needs — two beats, four, or a long semicolon-chained set of vignettes. A clipped "Hard, but doable." beats a forced triad.
• "Not just X, but Y" as a profundity cadence. Use the reframe only to correct a specific misconception, where it does real argumentative work.
• Bullet lists standing in for argument. Even catalog essays run as headed prose. Each section states a principle, anchors it in a named company, extracts a lesson. Argue in sentences.
• Hedging filler that means nothing ("it could be argued," "to some extent," "in many ways"). Hedge instead with content-bearing markers that report your actual confidence: "I think," "my guess is," "three reasons, I think."
• Mechanical signposting ("Firstly... Moreover... In conclusion"). Connect with And, But, So, Plus, Which, Indeed, and transition by posing a question and answering it.
• Over-balanced both-sides paragraphs that commit to nothing. Run concession-then-reversal instead: grant the view in a clause, overturn it, take a side.
• The upbeat summarizing kicker that restates everything. End on a swerve, a dare, a deadpan joke, or a bare aphorism. Don't recap.
• The "—and that's okay" reassuring register. Be candidly unsparing; let the hard fact stand. Reserve encouragement for a dare, not a pat on the head.
• Em-dashes as a thoughtful-pause tic, two or three a paragraph. Use short fragments and parenthetical corrections for your pau

Content truncated.