tl;dr
Rough voice notes become publishable writing in four steps: capture loose, let Auri Voice's cleanup pass strip the filler and false starts, then edit in three short passes, structure first, facts second, voice last. Cleanup hands you a readable transcript; the passes turn it into something you would publish under your name. This guide runs one real forty-second capture through the whole process, so you can see exactly what each step does and does not do.
The raw capture
Everything below starts from one capture: forty seconds spoken into a note, about whether to move a product launch. The capture rules fit in a sentence: finish your sentences, keep talking through mistakes, and correct yourself out loud in plain words. The speaking guide covers those habits; this one covers what happens after you stop talking.
You said: "okay so, um, thoughts on the launch date. marketing wants the twelfth but support only has, uh, two people that week. we could bring in a, no, wait. we tried contractors in March and onboarding took longer than the gig. so the twelfth is risky. the fifteenth, sorry, the sixteenth is a Monday, which support prefers. um, also App Store review, we should count three days for that. honestly the more I talk about this the more I think we just move it. we should move the launch to the sixteenth and tell marketing this week."
That take has everything a spoken draft collects: filler, an abandoned sentence, a date corrected mid-stream, and a conclusion that only shows up at the end because the speaker talked their way to it.
What cleanup fixed, and what it left
Auri Voice runs its cleanup pass before the text lands at your cursor. The app describes the pass in one line: it removes false starts and filler words, fixes punctuation, and follows spoken formatting commands. Here is what it made of that capture:
Auri typed: "Thoughts on the launch date. Marketing wants the twelfth, but support only has two people that week. We tried contractors in March, and onboarding took longer than the gig. So the twelfth is risky. The sixteenth is a Monday, which support prefers. Also, App Store review; we should count three days for that. The more I talk about this, the more I think we just move it. We should move the launch to the sixteenth and tell marketing this week."
The fillers are gone, the abandoned contractor sentence is gone, and "the fifteenth, sorry, the sixteenth" came out as the sixteenth. What cleanup did not touch is the shape. The conclusion still sits in the last sentence, the App Store point still floats in the middle, and nobody has checked that the sixteenth is a Monday. Cleanup fixes sentences; order, accuracy, and cuts are your job, which is what the three passes below are for.
| Cleanup handled | Still your job |
|---|---|
| Filler words | Paragraph order and the buried conclusion |
| False starts | Names, numbers, and claims |
| Spoken self-corrections | Repetition across paragraphs |
| Punctuation and capitalization | Cutting what is weak |
Pass one: structure
Structure first, always. Fixing sentences in a paragraph you are about to delete wastes the whole session, so on this pass read only for shape and do not touch wording.
- Read the draft once and write a one-line summary of each paragraph in the margin.
- Reorder those lines until the argument runs in the order a reader needs.
- Move the paragraphs to match, without rewriting them.
- Delete the first thirty seconds of the capture, which is almost always warm-up.
- Merge any two paragraphs that make the same point twice.
Spoken drafts bury the conclusion because you talk your way toward it. In the launch capture the whole point is the last sentence: move the launch to the sixteenth. Structure moves that sentence to the front, deletes "thoughts on the launch date" as warm-up, and pulls the App Store buffer next to the support problem, since both are reasons for the same decision. A minute of moving, no rewriting yet.
Pass two: facts
Facts get their own pass because they need a different kind of attention. Reading for accuracy while reading for flow means you catch neither, and fluent cleanup makes wrong details easier to miss.
- Check every name against the way that person or product spells it.
- Read numbers aloud digit by digit, including prices and counts.
- Confirm dates and weekdays agree with each other.
- Verify every claim you would not want to defend, or remove it.
- Retype identifiers rather than trusting a dictated URL or address.
Do this pass with the source open: the calendar, the spreadsheet, the meeting notes. For the launch note that means two checks. Is the sixteenth really a Monday, and does support really have two people that week? You said "the fifteenth, sorry, the sixteenth" while speaking; cleanup applied the correction, but only you can confirm the correction was right. A claim you cannot check in ten seconds gets softened into something you can support, or cut.
Pass three: voice
The last pass makes the draft sound like a person wrote it. Spoken sentences already carry rhythm, so the work is trimming rather than rewriting.
- Read the draft aloud. Every place you stumble marks a sentence to shorten.
- Cut hedges: kind of, sort of, I think, a bit, probably.
- Replace vague nouns with the specific thing you meant.
- Break any sentence longer than about thirty words.
- Cut the weakest fifth of the draft. Spoken drafts overexplain, because in conversation you keep talking until the other person nods.
Here is the launch note after all three passes:
The final version: "Let's move the launch to Monday the sixteenth. Marketing wants the twelfth, but support has only two people that week, and App Store review needs a three-day buffer. We tried covering a gap with contractors in March; onboarding took longer than the gig. I'll tell marketing this week unless someone objects by Friday."
Same facts, a third shorter, conclusion first. "The more I talk about this" and "honestly" did not survive the cut, and the deadline gained a date the spoken version never had. Additions like that Friday deadline only occur to you with the whole draft in front of you, which is why they belong in editing, not in capture.
A repeatable twenty-five minute session
Put the passes on a clock so the process survives a busy week. One block produces a finished short piece or a solid section of a longer one.
- Five minutes: capture by voice, loose, without stopping.
- Ten minutes: structure pass. Reorder, merge, delete the warm-up.
- Five minutes: facts pass with sources open.
- Five minutes: voice pass. Read aloud and cut what repeats.
Your raw material is already waiting when the block starts. On Mac, every dictation lands in History: press ⌘3 in Auri Voice, or use the search box when you remember a phrase but not the day. On iPhone, record into Smart Notes and the transcript stays with the note. Label each capture with a two-word topic and keep captures to one idea each, so the structure pass has clean pieces to move.
FAQ
How messy is too messy when capturing?
Almost nothing is too messy if you finish your sentences. Fillers, restarts, and spoken corrections get cleaned up. Half-abandoned sentences are the one thing worth avoiding.
How long should each editing pass take?
For a thousand-word draft, roughly ten minutes for structure, five for facts, and five for voice. Doing them separately is faster than one slow read.
Can I fix wording while the recording is still running?
You can, and it costs more than it saves. Speaking and editing use different attention, so switching every few seconds is slower than either task alone. Finish the take, then edit.
What do I cut first?
The first thirty seconds of the recording and every sentence that restates a point you already made. Spoken drafts warm up slowly and repeat.
Does this work for meeting notes too?
Yes, with a shorter version. Capture the decisions and owners, then run the facts pass immediately while names and numbers are still fresh.