The four modes
Voice-first writing works when each mode does one job. Capture generates material, structure orders it, polish fixes the language, and sending ends it. Trouble almost always comes from doing two at once.
- Capture: speak the whole thing, badly.
- Structure: reorder, cut repeats, mark what is missing.
- Polish: sentences, tone, grammar, names.
- Send: ship it and close the document.
The same four modes work for a two-line reply and a long brief. Only the time per mode changes, from seconds to an afternoon.
Capture: talk past the mistakes
In capture, the only measure of success is a complete rough draft. Restarting a sentence because you disliked the wording rebuilds the blank page you were trying to escape.
- Keep talking through a stumble and let the wrong words stand.
- Mark gaps out loud: “stat later”, “name to check”, “example goes here”.
- Ignore tone completely. Flat and finished beats elegant and half done.
- Stop when the idea exists on screen, not when it reads well.
Set a timer so capture has an edge to it. Modern dictation tools remove filler and false starts on their own, which is what makes this permission to ramble practical rather than just optimistic.
Structure: order before sentences
Structure is the pass people skip, and it is the one that saves the most time. Polishing paragraphs you are about to delete is the most common way to spend an hour and gain nothing.
- Read the draft silently and break it into sections.
- Move the real point to the top if it arrived halfway down, which it usually does when you talk.
- Delete the second and third versions of the same explanation.
- Fill each spoken placeholder or delete the sentence around it.
Do this in text, with your hands. Reordering by voice is slower than typing, and this is the one mode where the keyboard clearly wins.
Polish: one pass, two at most
Polish is where AI tools earn their place, on a draft that already holds together. Cap the number of passes, because a third one tends to trade real improvement for the feeling of progress.
- Run grammar and spelling across the whole draft at once.
- Use a paraphrase style when the facts are right and the tone is off.
- Translate only after the source draft is final.
- Check names, numbers, dates, and anything you promised by hand.
Auri Keyboard runs these in the field on iPhone with grammar help, 25+ paraphrase styles, and 20+ translation languages. On Mac, Auri Voice already cleans filler and punctuation during capture, so polish is usually one short pass on tone.
Send: the mode people skip
Sending is a mode, not an afterthought. A draft that sits in a folder cost you the whole workflow and returned nothing, and the fix is a rule you decide in advance rather than in the moment.
- Set the bar before you start: sendable, not perfect.
- One reread, out loud if it is important, then send.
- If you are stuck on a sentence, cut it. It is usually the one you were unsure about anyway.
- Close the document afterward so you stop reopening it.
Track how many drafts you actually send in a week. That number tells you more about the workflow than any measure of speed.
Which tool for which mode
The four modes stay the same across devices. The tools change, and so does which mode each device is good at.
| Mode | Mac | iPhone |
|---|---|---|
| Capture | One dictation shortcut into any app, local engine for private work | System mic or keyboard mic, wherever you are |
| Structure | Best place for it: keyboard, big screen, easy reordering | Workable for short pieces only |
| Polish | Editor plus optional AI on a finished draft | Keyboard tools in the field: grammar, paraphrase, translate |
| Send | Mail, docs, CMS | Messages, Mail, chat |
If you move between devices, keep one home document per piece. Drafts scattered across three note apps break the workflow more reliably than any tool limitation.
A day that survives interruptions
The workflow has to work on a day with three meetings and a delayed train. Short blocks in the right mode beat waiting for a free hour that never arrives.
- On the walk or commute: capture two replies and one outline on the phone.
- First desk block: structure the outline, cut repeats, mark missing facts.
- Before lunch: polish and send the email that has been waiting.
- Afternoon: a longer capture session on Mac, local engine if the topic is sensitive.
- End of day: one polish pass, then send whatever is ready.
When a block gets eaten, leave the draft in whatever mode it reached. A structured outline waiting for polish is a much easier restart than a blank page.
FAQ
Should I speak punctuation during capture?
Say it when the meaning depends on it, such as a question mark or a paragraph break. Do not chase perfect punctuation if it interrupts your pace. Cleanup tools handle most of it, and you fix the rest in the polish pass.
How long should a capture block be?
Five to ten minutes, ending when the whole idea exists in rough form. If you are still talking after fifteen, you are probably drafting two things and should split them.
Where do AI tools belong in this?
In cleanup and polish, after the draft has a beginning and an end. Running a rewrite on half an argument produces a well-worded fragment, which is harder to fix than a messy whole.
Can I do all four modes on iPhone?
For emails, messages, and short posts, yes. Long pieces are easier when you capture on the phone and structure on a bigger screen, because moving paragraphs is the one job a phone makes harder.
What if I lose the thread mid-capture?
Say a placeholder out loud, something like “come back to pricing”, and keep going. Search for those phrases during structure. Restarting the draft costs far more than the gap does.