Start with the built-in option in each tool. Claude Code has /voice, Cursor has Cmd+Shift+Space, and the ChatGPT apps have a microphone button. They’re free, they’re already there, and for one tool they’re usually enough. If you want one shortcut for all of them, plus your browser and email, you need system-wide dictation instead: Apple Dictation, or a push-to-talk app like SpokenKey.
We make SpokenKey, a small independent Mac app, so weigh our opinion accordingly. The facts about the other tools below come from their own documentation as of November 2026. We haven’t tested them hands-on.
Claude Code: the built-in /voice command
Run /voice in the Claude Code CLI. By default you hold Space to record and release it to stop. If you’d rather not hold a key, /voice tap switches to tap mode.
There are some conditions. You need to be signed in with a Claude.ai account, not an API key, Bedrock or Vertex. It doesn’t work in SSH sessions or cloud sessions. The audio streams to Anthropic’s servers for transcription, so it isn’t local. The default language is English, and about 20 are supported.
If you work on a remote box over SSH most of the day, that second limit matters. Voice input won’t be there when you need it, and a tool that runs on your own Mac is the way around it. Otherwise /voice is the cleanest way to dictate to Claude Code, because it’s built for that exact input box.
Cursor
Cursor has dictation in its chat input. Press Cmd+Shift+Space in the IDE to toggle it on and off. Since November 2026 it also has a separate realtime voice chat with agents, which is a different thing: you’re talking back and forth with an agent, not dictating a long prompt you can review. Both need a signed-in Cursor account.
For the long-prompt case, the dictation shortcut is the one you want. You speak, the words land in the chat box, you read them, and you send.
ChatGPT
The ChatGPT apps have a microphone button for dictating a message. OpenAI’s help page says the recorded audio "is sent to our models to be transcribed." So, same as Claude Code, it’s cloud transcription.
One caution: the Mac app was replaced by a new desktop app in July 2026, and we won’t describe its exact layout here. Look for the microphone button in the message box.
One shortcut across apps: system-wide dictation
The built-in options share a weakness. You learn three shortcuts, three sign-in requirements and three sets of limits, and none of them helps in your terminal, your issue tracker or your email.
System-wide dictation types into the focused field in most apps, so it’s the same habit everywhere you write. You have two routes on a Mac.
Apple Dictation
It’s free and built in. Turn it on in System Settings › Keyboard › Dictation. Then start it with the Microphone key, a shortcut such as "Press Fn (Function) Key Twice" or a custom one, or from Edit › Start Dictation. Stop it with Escape or the same key.
Apple says you can dictate text of any length without a timeout, but it stops automatically when no speech is detected for 30 seconds. That matters for prompts, because people pause to think. You’ll sometimes find it has stopped while you were deciding what to say next.
On privacy, the Keyboard settings show whether dictation is processed on your device and not sent to Siri servers. Apple’s privacy page says that otherwise what you say is sent to Apple. Apple’s own troubleshooting notes that you may need an internet connection. It works in many text fields, not all of them.
A push-to-talk app like SpokenKey
SpokenKey is a Mac app for Apple silicon (M1 or later) on macOS 15.0 or later. You hold Fn, speak, and release. While you talk, the partial words stay inside SpokenKey, so the field you’re typing in doesn’t change. When you let go, it finalizes the text on your Mac and inserts it into the focused field. Transcription runs locally, using a Whisper model bundled with the app. Your audio and transcripts aren’t sent to a server; the app goes online only to check your license and updates.
It costs $29 once, for one Mac, with no subscription and no account. There’s no free trial, but there is a 7-day refund. The menu-bar cue walks through "Recording · Local", "Finalizing…" and "Inserted · Not saved". Then you pick Save to History or Done. Push-to-Talk text isn’t saved unless you choose to save it.
Push-to-Talk needs two permissions, Input Monitoring so it can hear the Fn key and Accessibility so it can type into the field. If you see "Fn not enabled", turn on both in System Settings › Privacy & Security.
A note on terminals
SpokenKey inserts text only when it can verify the insertion into the focused field, and it doesn’t work in every app or field. Some terminal apps and editors may not allow that. If SpokenKey can’t insert into your terminal, it keeps the text and shows "Your text is ready to copy" with a Copy text button. You switch back to the terminal and paste.
That’s one extra step, not a failure, but it does change the feel. If your whole day is inside one terminal that won’t accept insertion, and you only dictate to Claude Code, /voice is the smoother choice. We can’t tell you which terminals work, so try yours before you buy. The 7-day refund covers that test, and our refund policy has the details.
How to speak a good prompt
Speaking a prompt isn’t like typing one. A few habits help.
Say file and function names slowly. A name like parseConfig is the kind of thing transcription gets wrong, and it’s easy to fix if you notice. For unusual identifiers, either spell them out or leave a gap and paste them afterward. Pasting is faster than correcting.
Talk through the goal first, then the constraints, then the "don’t touch" list. Something like: "I want the settings screen to save on blur instead of on a button press. Keep the existing validation. Don’t touch the API client." That order is how you’d brief a colleague, and it gives the model what it needs to stay in bounds.
Long, rambling prompts are fine. Models handle them well, and you’ll often give more context by speaking than you would have bothered to type. Don’t stop to edit mid-sentence. Finish the thought, then fix it.
Then read the text before you send it. Dictation makes small mistakes, and in a coding prompt one wrong filename can send an agent to the wrong place. Reading takes a moment.
What doesn’t work well by voice
Code symbols. Braces, brackets, backticks, underscores and camelCase are painful to dictate with any of the tools above, and we wouldn’t rely on any of them for it. Dictate intent, not syntax. Tell the model what you want and let it write the code.
SpokenKey also has no custom vocabulary, so it won’t learn your project’s names. It has one bundled model, no language picker (Whisper detects the language itself), and no AI rewriting or filler-word cleanup. What you say is what you get, with no editing pass. If you want your rambling tidied before it reaches the prompt box, SpokenKey isn’t built for that.
Questions
Can I dictate to Claude Code without signing in to Claude.ai?
Not with /voice, which requires a Claude.ai sign-in. A system-wide option like Apple Dictation or SpokenKey types into the terminal as text and doesn’t depend on that sign-in, as long as the terminal accepts it.
Does dictation work over SSH?
Claude Code’s /voice doesn’t work in SSH or cloud sessions. A system-wide tool runs on your Mac and types into your local terminal window (or, with SpokenKey, gives you the text to paste if it can’t insert), so it works whatever the terminal is connected to.
Is my audio sent to a server?
With /voice, audio streams to Anthropic’s servers, and ChatGPT’s help page says its audio is sent to OpenAI’s models. Apple Dictation depends on a setting you can check in Keyboard settings. SpokenKey transcribes on your Mac and doesn’t send audio or transcripts to a server.
Which one should I pick?
If you only dictate in one of these tools, use its built-in option. If you want one habit across your terminal, editor, browser and chat apps, try system-wide dictation, starting with Apple Dictation since it’s free. If you like holding a key and want local transcription, SpokenKey is $29 once. For a wider comparison, see our best dictation apps for Mac and SpokenKey vs Apple Dictation.