The habit that fixes most problems: focus first
Nearly every complaint about Mac voice typing traces back to one thing. The text goes where the cursor is, and the cursor was somewhere else. Recognition worked fine; the words landed in a search box or a background window.
Build the sequence into muscle memory before you tune anything else:
- Click into the field you want. Not the app, the field.
- Confirm you can see a blinking cursor.
- Press the shortcut, speak a full thought, stop.
- Read what landed before you keep going.
Two apps deserve extra care. In Slack, a thread pane can look active while the main channel box holds focus, so one mis-targeted message teaches the habit permanently. In browsers, clicking the tab is not the same as clicking inside the document. If insertion fails only in one app, click the field again and retry before you open settings, then check that the app still has Accessibility permission after its last update.
Mail: say the reply you would have typed
Email is where dictation pays off fastest, because most replies are three short moves: the answer, the reason, and the next step. Say those out loud in order and you have a complete draft in under a minute.
A pattern that works for almost any reply:
- Answer the question in the first sentence, including the word yes or no if that is the answer.
- Give one sentence of reason or context, no more.
- State the next step with a name and a date attached to it.
- Close with one short line.
Speak all of it in one take. The cleanup pass in Auri Voice removes the ums and the false starts and adds punctuation, so a messy delivery still arrives readable. What you check afterward is short: recipient names, numbers, dates, and any commitment you just made. Type email addresses and links by hand rather than dictating them, since a single wrong character in an address costs more than the seconds you saved.
Browser documents: prose by voice, structure by hand
Google Docs, Notion, and similar browser apps accept dictated text in the body without much trouble. The friction is everywhere else: headings, tables, comments, and suggestion mode all involve interface state that speech does not address well.
Split the work along that line:
- Create the skeleton first with the keyboard: headings, section order, any table shells.
- Click into a section and dictate its paragraphs as continuous prose.
- Leave comments and suggestions to the keyboard and trackpad.
- Fix links and inline code after the prose exists.
Watch two behaviors specific to the browser. Autosave needs the network even when local recognition does not, so on an unreliable connection dictate into a local file and paste in later. And collaborative cursors can move your insertion point when someone else edits the same block, so avoid dictating into a paragraph a colleague is actively rewriting. Both problems look like dictation bugs and are actually document sync.
Slack and chat: short, targeted, and read once
Chat rewards speed, which is exactly where voice helps, and it punishes mistakes publicly, which is why one extra second of care matters. The main risks are thread targeting and tone, not accuracy.
| Situation | What to do |
|---|---|
| Reply in a thread | Open the thread and click its compose box before speaking |
| Long explanation | Dictate it, then read it once for tone before sending |
| Code or a command | Type it. Never dictate a path or a flag |
| Bad news or pushback | Dictate the draft, wait ten seconds, reread, then send |
| Someone is typing in the same channel | Confirm your cursor is still in your box |
Spoken language runs warmer than typed language, which usually improves chat and occasionally overshoots. If a message includes a decision, a deadline, or a disagreement, read it aloud in your head once before pressing return. That single pass catches the sentence you would have phrased differently on a keyboard.
What cleanup handles and what you still own
Knowing the division of labor stops you from proofreading things that are already handled and from trusting things that are not. Auri Voice runs recognition and text improvement on device by default, so the mechanical mess is gone before you read the draft.
- Handled: filler words, repeated starts, self-corrections you spoke aloud, sentence punctuation, basic list and paragraph formatting.
- Yours: names, numbers, dates, links, commitments, and any word that is wrong while still being a real word.
- Yours: paragraph order and anything you said twice in different ways.
- Yours: tone for messages that carry consequences.
A thirty-second scan covers the second and third items on most messages. For anything client-facing, read the draft top to bottom once. The failure mode of good cleanup is that text reads so smoothly you skip the facts check, and a fluent sentence with the wrong figure in it is worse than an obviously rough one.
A working setup for a normal Mac day
Once the habits are in place, the setup is small. One shortcut, one microphone, and a short list of decisions you make once instead of every message.
- Pick a global shortcut you can press without looking, and keep it out of the way of app shortcuts.
- Choose one microphone and stop switching. Consistency beats quality here.
- Grant Microphone and Accessibility, then relaunch so insertion works everywhere.
- Add recurring names and product terms to your custom word list.
- Decide which apps you dictate into and which stay keyboard-only, such as your terminal.
Searchable history on the Mac covers the case where you dictated something good and the app ate it, which happens with browser documents more than you would like. Auri Voice keeps that history on your Mac. If you share a machine, review the history setting so your drafts do not sit there for the next person who logs in.
FAQ
Why does dictated text sometimes land in the wrong app?
Because a different window held focus when you pressed the shortcut. Click into the exact field first, including the specific Slack thread, then start dictation.
Can I dictate into Google Docs in a browser?
Yes. Insert generally works in the document body. Headings, comments, and suggestion mode are faster with the keyboard, so use voice for prose and hands for structure.
How do I dictate a bulleted list?
Say the list as sentences and let the cleanup pass format it, or create the bullet in the app first and dictate one item per bullet. Both work; the first is faster.
Should I dictate the whole email or one sentence at a time?
The whole thing. Long takes give the model more context and give you one editing pass instead of ten interruptions.
What still needs typing?
Email addresses, URLs, code, file paths, and anything with punctuation that matters more than wording. Speak the prose, type the strings.