How to use dictation in every Mac app
The practical guide to Super Voice Mode dictation: quick dictation, AI Writing, Live Transcription, and the small bits of polish that make spoken text land cleanly.
Super Voice Mode is built around a simple promise: speak naturally, and get text that looks like you wrote it. The app works anywhere normal Mac text input works — Slack, Gmail, Mail, Cursor, Xcode, Terminal, browsers, Obsidian, and whatever text field is in front of you.
There are three ways to use it. Quick dictation and AI Writing are free forever. Live Transcription is a Pro workflow, with 50 free Pro uses each week after your trial.
Quick dictation
Use this when you want to say a sentence, a reply, a note, or a quick block of text.
- Click where you want the text to go.
- Hold Right Option.
- Speak.
- Release.
Super Voice Mode records while the key is down, trims silence, transcribes locally, cleans up the text, and pastes it at your cursor.
That makes it feel less like opening a dictation app and more like typing with your voice. You stay in the app you were already using.
AI Writing
Use this when you want AI to write from an instruction, or when existing text needs to be transformed.
There are two entry points.
With highlighted text
Highlight text in any editable app, then hold your dictation key. By default, that is Right Option.
If the AI Writing setting is set to Talk to AI, Super Voice Mode treats the highlighted text as the thing you want changed. Speak the instruction, release, and the result replaces the selection.
Say:
- “Make this shorter.”
- “Rewrite this as a polite reply.”
- “Turn this into bullet points.”
- “Translate this to French.”
- “Make the tone more direct.”
If that setting is set to Replace, the dictation key behaves like plain dictation over the selected text instead. The dedicated AI Writing hotkey still opens AI Writing.
With no highlighted text
Use the AI Writing hotkey when you want the model to write from your instruction. By default, that is Shift + Right Option.
With no text selected, your spoken instruction is the source for the new text. Include the details you want written, and Super Voice Mode inserts the result at the cursor. Say:
- “Translate the following into French: I will send the proposal tomorrow.”
- “Write a friendly reply saying Tuesday works for me.”
- “Draft a short project update from these notes.”
- “Turn this idea into a three paragraph email.”
AI Writing uses the same local correction layer as dictation, but the job is different: instead of simply transcribing your words, it follows the instruction you spoke. If nothing is selected, it writes from that instruction at the cursor. If text is selected, it rewrites or transforms that text, depending on your AI Writing setting.
AI Writing is free.
Live Transcription
Use this when you are going to talk for a while: drafting an email, taking notes, thinking through a plan, or capturing a longer train of thought.
Double-tap the dictation key to start a hands-free session. Words stream to the screen with live subtitles, and you can stop when you are done.
Live Transcription is a Pro workflow. After your Pro trial, the free tier includes 50 Pro uses each week across Live Transcription, assistant turns, and spoken replies. Upgrade to Pro when you want those workflows unlimited.
If AI cleanup as you speak is enabled, Super Voice Mode can polish chunks as they arrive. Keep it off when you want the fastest, most literal capture, and turn it on when the extra cleanup matters. The important part is that the cursor keeps moving. If cleanup has trouble during a session, dictation keeps going with raw text rather than blocking your flow.
Smart formatting first, optional AI cleanup
The three modes are the main thing to learn. Underneath them, Super Voice Mode does a layer of smart formatting so you do not have to think about dictation commands all day. Optional local AI cleanup is there when you want more polish.
Most of that polish is deterministic: numbers, money, spacing, operators, filler removal, and punctuation cleanup. Optional local AI Correction can help with casing, sentence shape, and misheard words, but it is not the default tool for structured writing.
The user-facing version is simpler:
Keep default dictation fast and predictable. When you want something more specific — a list, an email, a translation, a summary, or a tone change — use AI Writing and say the instruction directly.
Numbers, dates, times, and money
These are designed to feel invisible. You say the human version; Super Voice Mode writes the useful version.
The goal is not to memorize a command language. The goal is that common spoken shapes become the text you expected.
What gets cleaned up
Technical text
Dictation gets more useful when it can handle the bits people usually give up and type by hand.
Super Voice Mode can convert spoken operators and assemble common technical shapes:
- “underscore” for
_ - “slash” for
/ - “dot” for
. - “no space” when you need words joined
- email-style phrases like “alex at example dot com”
- path-style phrases like “slash users slash mike slash desktop”
This is especially useful in Cursor, Xcode, Terminal, GitHub, Linear, Slack, and docs where names, commands, file paths, and code-adjacent text show up constantly.
New lines, paragraphs, and capitalization
If you want more control, you can enable spoken formatting commands in the Dashboard.
Useful phrases include:
- “new line”
- “new paragraph”
- “cap” plus the next word
- “all caps”
- “no caps”
Keep these as tools, not habits. Most everyday dictation works better when you speak normally and let AI Correction handle the written shape.
Fix names and jargon with Dictionary
Some words are personal. Names, project codenames, company jargon, internal acronyms, and weird product names may be misheard the same way every time.
That is what the Dictionary is for.
Use it for the text you say all the time but speech recognition does not reliably know: your name, your company name, customer names, product spellings, email addresses, LinkedIn profiles, commands, file paths, stored prompts, and snippets you do not want to dictate from scratch.
Practical examples
Add entries like:
get hub -> GitHub
kube control -> kubectl
alex email -> [email protected]
alex linkedin -> https://www.linkedin.com/in/alex-example
voice mode repo -> /Users/mike/Projects/voicemode
linkedin outreach prompt -> Write a concise LinkedIn connection note.
Mention the product launch, keep it warm,
and avoid sounding automated.
my email signature -> Alex Example
Product Lead
[email protected]Dictionary entries run before optional AI Correction, so the correction layer sees the fixed version. It is predictable, fast, and good for words that speech-to-text models do not know.
Good Dictionary entries are usually short spoken phrases that expand into exact written text:
- People and companies: “mike h” ->
Mike Haydon, “acme labs” ->Acme Labs - Product names: “super voice mode” ->
Super Voice Mode, “codex cli” ->Codex CLI - Email and handles: “alex email” ->
[email protected], “support email” ->[email protected] - Profiles and links: “alex linkedin” ->
https://www.linkedin.com/in/alex-example - Code and commands: “kube control” ->
kubectl, “npm run dev” ->npm run dev - Prompts and snippets: “linkedin outreach prompt” -> a saved prompt you reuse, “my sign off” -> your usual closing
- Paths: “voice mode repo” ->
/Users/mike/Projects/voicemode
The trick is to make the spoken side natural. Say what you would actually say out loud, then map it to the exact text you want on screen.
Undo and send
If the last dictation went somewhere wrong, say:
scratch thatSuper Voice Mode can undo the previous dictation.
You can also configure submit phrases. If you end a dictation with one of those phrases, Super Voice Mode pastes the text and presses Return. That is useful in chat apps where you often want to speak a message and send it in one motion.
App-specific tips
Slack, Messages, and chat apps
Short dictations work best. Say one message, release, check it, send. If you use submit phrases, choose a phrase you would not say accidentally.
Gmail and Mail
Use quick dictation for paragraphs and AI Writing to draft, translate, or tighten before sending. With no selection, include the reply details in what you say: “write a friendly reply saying Tuesday works for me.” With text selected, “make this warmer” or “make this more direct” works better than trying to re-dictate the whole email.
Cursor, Xcode, and code editors
Use dictation for comments, commit messages, issue notes, and prose around code. For symbols, file paths, and command names, the Dictionary is your friend. Add the few names you say every day instead of fighting the model every time.
Terminal
Be deliberate. Terminal text often executes things, so keep commands short and check before pressing Return. Terminal output mode keeps pasted text plain.
Browsers and docs
Click into the field first, then dictate. If a web app has unusual selection behavior, use AI Writing without a selection to write from your spoken instruction at the cursor, or use it on native text fields instead of canvas-based editors.
A good default setup
Start with:
- Right Option for quick dictation.
- Talk to AI for the highlighted-text dictation key behavior.
- Shift + Right Option as the dedicated AI Writing hotkey.
- AI Correction off by default, unless you specifically want extra sentence polish.
- Spoken numbers and operators on.
- Formatting and capitalization commands off until you need them.
- A few Dictionary entries for your own names and jargon.
Then use it for a day. The first things to customize are usually not the big settings. They are the words you say constantly that your Mac has never seen written down.
What stays local
Dictation runs on your Mac. Speech-to-text is local. AI Correction is local. The Dictionary is local.
The assistant is different: if you choose Claude, Codex, Gemini, or another external backend, the text you send to that assistant goes to that provider. Dictation itself does not need that.
For most people, the clean mental model is:
- Dictation writes for you.
- AI Writing writes or transforms for you.
- Live Transcription captures longer thoughts.
- The assistant answers or acts when you explicitly ask it to.