The most boring best man speech you ever sat through was written by someone who loved the groom more than anyone else in that room. That is the quiet tragedy of bad speeches: they almost never fail from a lack of feeling. They fail because they were composed in a language the mouth doesn't speak — drafted in silence, at a keyboard, in long elegant sentences that read beautifully on a laptop at midnight and die the moment they hit open air. The speech didn't fall apart at the podium. It fell apart in the document, weeks earlier, and nobody noticed until there was a microphone involved.

If you have a toast, a eulogy, a wedding speech, or a conference talk coming up, the single most useful thing you can learn is this: written English and spoken English are not the same language. And you are almost certainly drafting in the wrong one.

Written English and spoken English are two different languages

Linguists who study the difference — most famously Wallace Chafe, who spent years comparing transcripts of natural speech with written prose — found that the two forms are structurally distinct, not just stylistically. Spoken language arrives in short bursts, which Chafe called intonation units: a handful of words at a time, roughly one new idea per unit, shaped by the breath. Speech leans on simple clauses strung together with and, but, so. It uses verbs. It addresses the listener directly. It repeats itself, on purpose, because repetition is how ears keep their footing.

Written language does the opposite. Freed from breath and real time, it stacks clauses inside clauses. It compresses actions into abstract nouns — the celebration of their commitment instead of we're here because they chose each other. It buries the verb three lines from the subject and trusts the reader to hold everything in place.

On the page, that density is a feature. Out loud, it's a demolition. When you type your speech, you inevitably produce written English — that's what keyboards summon. Then you stand up in front of a hundred people and try to pronounce a paragraph. Your lungs weren't consulted during drafting, so the breaths land in the wrong places. Your eyes lock onto the page because the sentences are too architected to hold in your head. Out comes the flat, careful cadence everyone recognizes instantly: the reading voice. The voice of someone reciting a document at people instead of talking to them.

Your audience can't re-read

Here's the part speakers forget: reading is interactive and listening is not. A reader controls the tape. Eyes backtrack constantly; when a sentence gets tangled, the reader loops to the start of the clause and rebuilds it. Research on eye movements in reading shows regressions — jumps back to earlier text — are a routine part of comprehension, not a failure of it.

A listener gets no rewind. Speech is a live stream, and the buffer is working memory — famously small, famously fleeting. Psycholinguists have long observed that deeply embedded sentences, the kind written prose produces effortlessly, are brutal to parse by ear: the listener has to hold the subject suspended while clause after clause piles in, waiting for a verb that arrives after the mental shelf has already collapsed. A sentence that reads as sophisticated is, acoustically, a memory test your audience will fail. And when they fail it, they don't ask you to repeat yourself. They just quietly stop following, smile politely, and start thinking about the buffet.

This is why professional speechwriters talk about writing for the ear as a separate craft. Peggy Noonan, who wrote for Reagan, built a whole book — On Speaking Well — around the idea that a speech is not an essay delivered standing up. The ear needs shorter units, more signposting, more repetition, and more direct address than the eye would ever tolerate. What an editor would cut as redundant, a listener experiences as a handrail.

Draft with your mouth, not your fingers

So here is the method, and it feels almost too simple: don't start by writing your speech. Start by saying it.

Before you open a document, talk the speech out — as if you were telling the story to one friend across a dinner table. Tell them how you met the groom. Tell them what your mother was actually like, not what eulogies say mothers are like. Record it. Then transcribe it and look at what you have.

What you'll find is that your transcript already has the qualities speech coaches spend sessions trying to install. The sentences are short because your breath made them short. The verbs are strong because speech doesn't nominalize. The warmth is there because you were talking to someone, and English spoken at a person carries a fundamentally different temperature than English typed at a screen. Yes, the transcript is messy — there are false starts and detours. But mess is cheap to fix. Stiffness is nearly impossible to fix, because it's baked into the grammar itself. Editing spoken bones into a clean speech takes an evening. Performing written prose into warmth takes an acting degree.

Then edit on the page — cut the rambles, reorder the stories, sharpen the landing lines — but protect the spoken skeleton. Every time you're tempted to "upgrade" a sentence into something more impressive, read the upgrade aloud. If you stumble, or run out of air, or hear the reading voice creep in, the written language is recolonizing your draft. Two edits matter more than all others: the one-breath rule (if you can't say a sentence comfortably in one breath, it's two sentences) and the verb hunt (every -tion, -ment, and -ness is a small verb that someone flattened — set it back on its feet and give it a human subject).

None of this is a modern trick. For most of history, speeches were composed this way by necessity — rhetoric was an oral art long before it was a written one. The keyboard is the anomaly. You're not learning a hack; you're removing one.

Your next moves

  • Record before you write. Set a timer for three minutes and tell the story of your speech — the couple, the person, the idea — out loud, as if to a friend at dinner. Do it before any document exists. Transcribe it. That transcript is draft one.
  • Do a stumble pass. Read your current draft aloud, pen in hand, and mark every spot where you trip, gasp for breath, or hear your voice go flat. Split every marked sentence in two. No exceptions.
  • Run the verb hunt. Search your draft for words ending in -tion, -ment, and -ness. Rewrite each phrase so a person is doing something: not "the continuation of their friendship," but "they stayed friends."
  • Test the opening on one human. Say your first thirty seconds — from memory, not the page — to one real person. If you can't hold it in your head after two tries, it's still written English. Simplify until it sticks.
  • Rehearse standing up, out loud, twice. To an empty chair, at full volume, pauses included. A speech rehearsed silently in your head has never actually existed yet.

Where Quill fits in

The speak-first method has one point of friction: getting your spoken draft into text you can actually shape. That's the entire reason Quill exists. Open any app — your notes, your doc, an email to yourself — talk your toast through the way you'd tell it to a friend, and watch it land as clean text in real time, with the false starts already smoothed away. Everything happens on your device, which matters when you're drafting a eulogy or a wedding speech full of things you've never said out loud to anyone; those words never touch a server. And when a paragraph needs tightening, one tap rewrites it — while you guard the spoken bones that make it sound like you. If you've got a speech coming and a blank page glaring at you, try talking instead at quill.lumenlabs.works.