tl;dr
Dictating into Mail, browser documents, and Slack on a Mac comes down to three habits: click into the field before you speak, say the whole message in one take, and check names and numbers before you send. Auri Voice inserts at the cursor from a global shortcut, and its cleanup pass strips filler, false starts, and punctuation slips before the text lands, so what arrives already reads like writing. Formatting, links, and code stay manual work.
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. In browsers, clicking the tab is not the same as clicking inside the document. If insertion fails in only 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.
If Auri Voice is not set up yet, the shortcut guide covers installation, both permissions, and troubleshooting; the rest of this guide assumes the shortcut already works.
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, mess included:
You said: "yes we can um we can ship the report Friday, the data cleanup took longer than we thought, so I'll send a draft Thursday no wait Thursday morning and the final Friday by noon, thanks"
Auri typed: "Yes, we can ship the report Friday. The data cleanup took longer than we thought, so I'll send a draft Thursday morning and the final Friday by noon. Thanks."
The cleanup pass removed the filler, kept the spoken correction, and split the run-on into sentences. What you check afterward is short: recipient names, numbers, dates, and any commitment you just made. Type email addresses and links by hand, since a single wrong character in an address costs more than the seconds you saved. And when Auri keeps mishearing a client or colleague name, add it once under Settings > Dictation > Text > Vocabulary ("Add a word, name, or phrase…"); the custom words guide covers what belongs on that list.
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 come from document sync, not dictation.
Browser documents also lose drafts. When a reload swallows a paragraph you dictated, the take is still in Auri Voice's History: press ⌘3 in the app, or open Recent from the menu bar icon. If you share the Mac, Settings > Dictation > History is where you turn saving off or delete everything.
Slack and chat: short, targeted, and read once
Chat is where voice is fastest, and also where a mistake is public. 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 |
Dictated messages come out more casual than typed ones, because you are talking. In most chat that reads friendly. In a message that carries a decision, a deadline, or a disagreement, it can read too loose, so reread that one before you press return. That single pass catches the sentence you would have phrased differently on a keyboard.
What cleanup handles and what you still own
Auri Voice's cleanup pass removes false starts and filler words, fixes punctuation, and follows spoken formatting commands such as "new line" and "bullet point", so the mechanical mess is gone before you read the draft. Knowing that division of labor saves you from proofreading what is already handled, and from trusting what is not.
- 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.
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 "bullet point" before each item and the cleanup pass formats the list, or create the bullets 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 where a single character matters. Type those by hand and dictate the sentences around them.