Writing a first draft by voice
· 6 min read
Dictating a first draft works for a reason that has nothing to do with typing speed: it is very hard to edit while you are talking. You say the thing, the words land as text, and the fixing happens later in a separate pass with your hands. What you need is somewhere disposable to put the words, a plan of what you are going to say, and the discipline not to stop when a sentence comes out wrong.
Why the first draft is the part that suits voice
Writing stacks two jobs on top of each other: deciding what to say, and choosing the words. Typing lets you do both at once, which sounds efficient and is why so many drafts stall in the second paragraph. You write eleven words, dislike three of them, rewrite the sentence, and lose the thought behind it.
Speech does not permit that. You cannot back up neatly, so you keep going, and at the end you have a whole ugly draft instead of a beautiful third of one. Editing an ugly draft is ordinary work. Starting one is not.
The corollary matters as much: a dictated draft is a draft. Nobody should read it but you, and it will contain sentences that make no sense. That is the deal.
Decide what you are going to say first
Five minutes here saves the whole session. Before you open a microphone, write a skeleton with your hands: five to nine lines, one per section, each stating what that section does rather than what it is called.
Something like: “explain why the invoice is late”, “give the two dates that matter”, “say what we are doing”, “ask for the extension”. These are cues, not headings. A paragraph of prose makes a poor cue, because you will read it aloud instead of talking.
Then speak to one line at a time. When a line is spoken, glance at the next. This is the entire method, and it is the difference between dictating for twenty minutes and wandering for twenty minutes.
Put the words somewhere disposable
Not the email you are going to send. Not the document with the tracked changes and the client’s name on it. Open a plain document in Notes or TextEdit and treat it as scrap paper.
Two reasons. You will say things you would rather were not in a real file, and rich editors on the web do strange things with large blocks of inserted text, particularly around lists. Fighting a formatting engine while holding a thought is how a session ends early.
Move the text into its real home in the cleanup pass, when it is text you have read.
Set up for a long stretch, not a sentence
- Open System Settings, Keyboard, and turn on Dictation if it is not already on. Set the language and choose a shortcut you can reach without looking down.
- Set Microphone Source explicitly if you have anything better than the built-in microphone. A headset boom that sits near the corner of your mouth beats a laptop microphone across a desk by a wide margin.
- Turn Auto-punctuation on for prose. For a first draft, inferred commas and full stops are better than saying “comma” three hundred times.
- Put water within reach and close the door. Both matter more than they sound.
- Dictate two paragraphs as a test and read them back. If accuracy is poor, fix the microphone before anything else.
The built-in shortcut is a toggle, and it stops on its own after a stretch of silence, which for long-form drafting means restarting it every time you pause to think. If you find yourself doing that constantly, it is the strongest sign that a held key would suit you better.
Speak in paragraphs and refuse to stop
The technique is simple to describe and takes a session or two to get used to.
Say a whole paragraph before you look at the screen. Watching the words appear invites correction, and correction is the thing you are trying to postpone.
Leave the mistakes. A wrong word is a search-and-replace later. A lost thought is gone.
Mark holes out loud with a word you would never otherwise use, then move on. Say “flagpoint” where you need a statistic, a name or a link, and search for it in the cleanup pass. Trying to remember what to look up is a worse use of your attention than saying a silly word.
Talk to somebody. Imagine one specific reader and address them. Dictation aimed at nobody reads like a memo; aimed at a colleague, it reads like writing.
Stop while you still have something to say. Sessions that run past the plan produce the paragraphs you cut first.
The cleanup pass
Do this later, ideally not the same hour, and always with your hands.
Start with structure: reorder the paragraphs into the order the reader needs before touching a sentence. Dictation produces material in the order it occurred to you, which is not the same thing.
Then read the draft on screen and cut. Spoken language repeats itself, restates the point at the end of a paragraph, and reaches for “sort of” and “basically”. Searching for your own filler words is quick and slightly humbling; most people have three, and the same three every time.
Fix the misheard words last. They are obvious, they are usually names, and they are the least important problem in the file.
Some dictation apps do part of this for you. Voice Keyboard Pro has an optional Smart Rewrite that strips filler words and shapes tone, a personal vocabulary so names and project codes stop arriving mangled, and a held key (Left Control by default) that suits the speak-then-think rhythm better than a toggle, with text arriving 200 to 500 ms after you release. It needs macOS 14 Sonoma or later. The free tier is 5 dictations a day, enough to see whether the method suits you and not enough for a drafting session; unlimited use is a subscription covering the Mac and iPhone together with a 7-day trial. Use automatic cleanup with your eyes open: a rewrite that tidies your voice out of the draft has not helped.
What voice does not fix
It does not make you a better thinker. A draft spoken from a bad plan is a bad draft delivered sooner, and the speed can disguise that for a surprisingly long time.
It does not suit everything either. Anything with a lot of structure on the page (tables, code, heavily formatted documents) is faster to type, and short precise text such as a subject line or a commit summary is usually quicker with fingers than with a microphone and a correction. Dictating into Slack, mail and a code editor on a Mac covers where the line falls in each of those.
And it needs a room. Open-plan drafting out loud is a real constraint, not a small one, which is why a lot of people who draft by voice do it early, at home, or in a car.
Questions
How many words an hour should I expect? Most people speak far faster than they type and then give a chunk of it back in editing, so measure the finished draft rather than the raw transcript. Do the same task both ways once and use your own numbers.
Is it worth dictating if I am a fast typist? Sometimes. The gain for fast typists is usually not speed but the separation of drafting from editing, and the fact that you can do it standing up or walking. If you type quickly and finish drafts easily, there is less here for you.
Should I dictate on an iPhone instead? It is a good option away from a desk, and a phone held near your face often hears better than a laptop across a table. The awkward part is moving the text back, so pick a destination that syncs to the Mac before you start rather than after.