Speech is faster than typing and far less careful. That combination is useful at exactly one stage: the pass where getting material out matters more than getting it right. What follows covers when to talk instead of type, how to dictate something you can actually work with, and what to repair in the transcript afterwards.
- Dictate when you know what to say and cannot type it down.
- Carry three or four points; never talk into a blank room.
- Speak the structure aloud: new paragraph, example, note to self.
- Cut, order, then tighten; polishing before cutting wastes the work.
- Check names, numbers, and terms, where transcription fails quietly.
When speaking beats typing
Dictation wins when you already know roughly what to say and cannot get it down, which makes it a fast way to turn notes into a draft. Explaining a section aloud to an imaginary colleague tends to produce the plain version you have been failing to type. It also wins on a walk or a commute, where the alternative is not typing but nothing.
It loses when you do not yet know what you think. Speech has no backspace and no scroll, so it cannot hold an argument still while you rearrange it. Dictating into uncertainty produces long, warm, circular audio that is genuinely painful to edit later.
Talk to a spine, not to a blank room
Put three or four points on a card before you press record, so the transcript arrives as material your notebook can hold rather than an unbroken block. Not a full plan, just enough that each time you finish a thought you know where to go next. Without it, most people talk in one long spiral and arrive back at the same idea four times over.
Say the point of the section out loud first, as a single sentence. That sentence orients everything after it, and it often survives into the finished piece unchanged. If you cannot say it, the section is not ready to be dictated and no amount of talking will fix that.
A transcript is not a draft, but it is a much better problem than an empty page.
How to dictate a usable pass
Speak in whole sentences and let each one finish. Resist restarting a sentence halfway through, because a transcript full of false starts takes longer to clean than the restart ever saved. If a sentence goes wrong, finish it badly and then say the next one properly.
Say your structural markers aloud: new paragraph, this is the example, note to self, come back to this. Spoken labels survive transcription and become the handles you edit by. Twenty minutes is a sensible ceiling per recording, since a shorter file is much easier to reuse.
Wherever you record, the audio needs to land where the writing happens. Strut takes a voice note from its notes panel, or an audio file you upload, transcribes it, and files the result as a rough note with a short list of writing leads underneath the text.
What dictated prose reliably gets wrong
Four faults appear in almost every transcript. Sentences run long and join themselves with and or so. Filler survives: sort of, basically, I think. Pronouns lose their referents, because you were pointing at something in your head. Emphasis vanishes, since your voice carried it and text cannot.
A fifth is subtler. Speech repeats itself to keep an audience with you, which is right in a room and wrong on a page. Expect to delete the same point stated three ways, and to keep whichever version came out shortest rather than the one you liked saying.
Turning a transcript into a draft
Read the transcript once and change nothing, marking the parts that are alive. Dictation usually produces two or three passages better than anything you type, surrounded by material that is merely warm. Your job is to find those and build outward from them.
Then work in three passes: cut, order, tighten. Delete filler and repetition first, since it can account for a third of the words. Move the surviving blocks into a sensible sequence. Fix prose last, because polishing sentences you are about to delete is wasted effort.
Punctuation, names, and the cleanup tax
Transcription handles ordinary speech well and specialist vocabulary badly. Proper nouns, product names, technical terms, and figures are where errors cluster, and they are the errors that survive all the way to publication, because they arrive as confident, fluent text that reads like everything around it. Nothing looks wrong.
Spell an awkward name aloud the first time you use it, and say a number twice if it matters. Accept that some punctuation will be wrong, because pauses are not commas. Budget the cleanup honestly: a dictated pass is fast to produce and not free to finish.
Where dictation does not belong
Do not dictate anything that must be exact on the first attempt. Quotations, figures, legal wording, and anything published without a careful edit belong in typing, where you watch the characters as you commit them. Voice is for volume, not for precision.
It also suits some writers badly, and that is not a discipline problem. If you compose by watching sentences form on screen, speech removes the very thing that makes writing work for you. Try it on one difficult section, judge the transcript, and stop if the answer is no.
Two minutes of walking speech, and the paragraph inside it
right so the point of this section, um, is that most training programmes fail in the first month and not because the training is bad, it's, the training is usually fine actually, it's that nobody changes the job afterwards, so you go back to your desk on the Monday and everything, all the incentives, all the deadlines, they're exactly the same as they were before you left, and so obviously you do what you always did, that's not, that isn't a motivation problem, it's a, sorry, new paragraph, it's a design problem, and I keep saying that out loud but I don't think I've written it down anywhere. note to self, find the example with the hospital rota
Most training programmes fail in the first month, and rarely because the training was poor. The training is usually fine. What fails is the return. On Monday the desk, the deadlines, and the incentives are exactly as they were, so people do exactly what they did before. That is not a motivation problem. It is a design problem. Cut from the transcript: three false starts, one apology, and the same claim stated twice. Kept: the Monday image, which came out better spoken than anything typed for this section all week. Carried forward as a marker: the hospital rota example, which the recording asked for and the draft does not yet have.
The recording ran under two minutes and yielded one paragraph. That ratio is normal, and the paragraph arrived during a walk that would otherwise have produced nothing.
Common mistakes
| The mistake | Why it costs you, and what to do instead |
|---|---|
| Dictating before you know the point | Speech cannot hold an argument still while you rearrange it, so uncertainty comes out as circling. Spend five minutes naming what the section argues, then record. If the point will not come, the recorder is the wrong tool. |
| Editing the transcript line by line | Repairing sentence one before you know whether it survives is the slowest route through. Read the whole transcript first, mark what is alive, delete the rest in one sweep, and only then start writing. |
| Treating the transcript as a draft | It is raw material wearing a false air of completeness, because it is fluent and long. Sent to a reader after light tidying, it sounds spoken and argues badly. Give it a real rewrite before anyone else sees it. |
| Skimming past names and numbers | Transcription errors in proper nouns and figures arrive as well-formed text, so proofreading glides over them. Check every name, number, and technical term against a source before the passage moves anywhere near a draft. |
Your checklist
- Write three or four points on a card before recording.
- Say the section's point aloud as your first sentence.
- Speak structural markers: new paragraph, example, note to self.
- Keep each recording under twenty minutes.
- Read the transcript once, marking the passages worth keeping.
- Cut, reorder, tighten, then verify names and numbers.
Key terms
- Raw pass
- A first spoken version of a section, recorded for volume rather than accuracy, and expected to lose a large share of its words before it becomes prose.
- Structural marker
- A phrase spoken into a recording, such as new paragraph or note to self, that survives transcription and gives the editor a handle to work with.
- False start
- An abandoned sentence the speaker restarts mid-clause, unnoticed in conversation but preserved in a transcript, where it is the most common thing to delete.
Turning a pile of notes into something?
Frequently asked questions
Is dictation faster than typing?
+
For getting words out, usually. For producing a finished paragraph, often not, because the cleanup is real work. The gain is largest when the alternative is no writing at all: a commute, a walk, or an evening when sitting at a screen is out of the question.
How do I stop it sounding like a transcript?
+
Speak in finished sentences, then read the result aloud. Passages that still sound spoken usually break one of two rules: they run past two clauses, or they restate a point already made. Fix those two habits and most of the problem disappears on its own.
Can I dictate a whole chapter?
+
You can dictate its raw pass, in segments rather than one long recording. Do the structural thinking first. A chapter dictated without a spine takes longer to assemble than it took to speak, and it frequently has to be recorded again.
What about noisy places?
+
Wind, cafés, and speakerphone distance all degrade a transcript, and the damage lands on precisely the words you care about. Get the microphone close to your mouth. A slightly awkward spot nearby beats a comfortable one across the room every time.
Should I dictate revisions too?
+
Revision notes, yes. Talking through what is wrong with a section is often clearer than annotating it, and the recording doubles as a plan. Revised prose, rarely. Rewriting is a precision job, and precision is the thing speech gives away.
What if I hate hearing my own recordings?
+
Almost everyone does, and you never have to listen. Work from the transcript alone and delete the audio once the text is safe somewhere else. The recording is a delivery mechanism for words, not a performance you are obliged to sit through and assess.