How to dictate punctuation

Why dictated text comes out as one long breath

A speech engine hears sound and writes words. It does not hear the silence where you'd normally put a comma, and it cannot see the period at the end of your thought. So unless you tell it otherwise, it gives you exactly what it heard: a wall of words with no marks. "okay so review the doc send it to maria and remind me thursday" — that's the raw shape of dictation, and it's the same on every tool.

There are two honest fixes, and the rest of this guide is just those two. You can speak the punctuation yourself, naming each mark as you go, which both Windows and macOS support out of the box. Or you can let software add the marks afterward — the operating systems do a basic version of this, and a dedicated tool can run a full cleanup pass that fixes the punctuation along with the filler words.

Most people start with the spoken commands because they're built in and they work the moment you turn dictation on. So that's where we'll start too. Then I'll show you the part that made me stop saying "period" out loud, which I'd been doing for the better part of a year like some kind of court stenographer.

Say the mark and it lands as the symbol

The spoken-command way is simple once you've done it once. While dictating, you say the name of the punctuation mark, and the engine writes the symbol instead of the word. Say "Hello comma how are you question mark" and you get "Hello, how are you?" The trick is that the mark goes exactly where you said it, so you're placing punctuation in real time as you talk.

This works the same whether you use a built-in dictation tool or a system-wide hotkey that pastes at your cursor. Press a key, talk, name your marks, release. A small capsule shows up while you speak so you know it's listening, and the text lands wherever your cursor is — an email, a doc, a chat box, all the same to it:

The one habit to build is saying the mark without a pause before it. Speak "the meeting is Thursday period" as a single phrase, not "the meeting is Thursday … period," or some engines write the word "period" instead of the dot. It feels unnatural for about a day, then your brain files it away, the same way it eventually stopped fighting the hotkey for dictation on Windows. After that you stop thinking about it. The full list of which words map to which marks is two sections down.

Let an AI pass add the punctuation instead

The other path skips the spoken commands entirely. You talk in plain sentences, and an AI cleanup pass reads the transcript and adds the commas, periods, and paragraph breaks for you. With Whisper this is a built-in step you trigger by voice. You need a Mac on Apple Silicon or a Windows 10-or-newer PC, a working microphone, and a couple of minutes. The whole local pipeline is free for any signed-in account, with no payment method asked for at sign-up. Here's the sequence.

That's the whole pitch for the automatic route: you stop being a stenographer and go back to being a person who's just talking. The catch — and there's always a catch — is that the AI decides where the marks go, not you. For most prose that's fine. For the cases where it isn't, the manual commands are right there, and the last section covers exactly when to use them.

The spoken punctuation commands that actually work

Here are the commands I'm confident about, because they're documented by Microsoft and Apple and they behave the same across both. Say "comma" for a comma, "period" or "full stop" for a period, and "question mark" for a question mark. For breaks, "new line" moves the cursor down one line — like pressing Return once — and "new paragraph" starts a fresh paragraph, like pressing Return twice. Those five carry the vast majority of everyday writing.

A few honest caveats, because this is where the listicles oversell. Both operating systems support a much longer list of marks — exclamation point, colon, quotes, dashes — but I'd point you to Microsoft's and Apple's own command pages for the full set rather than reciting a list from memory that might drift between OS versions. The command words can also vary slightly by language and region. If you dictate in something other than US English, check your OS's dictation help for the localized command words, because "period" in English is not the command in French.

The other thing worth knowing: both Windows Voice Typing and macOS Dictation now have an automatic-punctuation setting that adds basic commas and periods for you while you speak, no commands needed. On Mac it's under System Settings, then Keyboard, then Dictation, where you can toggle Auto-punctuation. On Windows, voice typing has an automatic-punctuation option in its settings too. So even the built-in tools give you a taste of the hands-off approach. The dedicated AI pass goes further — it also fixes filler words and run-ons — but for a quick note, flipping the OS auto-punctuation switch might be all you need.

Local or cloud: which mode handles the cleanup

If you go the AI-punctuation route, the next question is where the work happens. The transcription and the cleanup pass can both run fully on your own machine, or in the cloud through your own OpenAI key. For most text — emails, notes, drafts — I'd try local first. It runs offline, nothing leaves your laptop, and a private message about your kid's school or your boss's budget has no business sitting in a vendor's logs just because you wanted punctuation. Here's how the three paths differ, because the app makes you pick and I'd rather you pick well:

The three options, plainly:

• Local Parakeet — NVIDIA's TDT engine, around 600 MB, and the fastest local option — 5 to 10 times faster than Whisper on CPU. Covers English plus 24 other European languages, 25 in total. No translate-to-English. If you write in English or another European language, this is the quick, fully offline pick.
• Local Whisper — slower than Parakeet on the same machine, but the multilingual builds cover 99 languages and can translate to English. The English-only builds are English-only, not 99. Pick this for Chinese, Japanese, Korean, or any translation work, which Parakeet can't do. Default English model is around 480 MB.
• Cloud (OpenAI, BYOK) — best accuracy and web access, using your own OpenAI key billed straight by OpenAI. Transcription runs on gpt-4o-mini-transcribe by default. Needs internet, so it's the one path that leaves your machine. The Cloud surface is part of Whisper Pro.

For the AI cleanup specifically, the local pass runs through Ollama on your own machine, and the cloud pass uses gpt-5-mini by default. The boring truth is that for adding commas and periods to ordinary writing, local handles it without complaint. Cloud earns its place when you want top-tier accuracy on a hard recording or need the model to pull a fact off the web mid-sentence. For punctuation alone, start local.

What the AI cleanup pass actually does to your text

This is the part that made me stop saying "comma" out loud. Raw dictation comes out as a run-on — you say "okay so review the architecture doc tag it project alpha and remind me thursday um before the standup," and that's the unpunctuated wall any speech engine hands you. The cleanup pass reads that and puts it back together as something you'd actually send.

With Whisper you trigger it by leading with the activation phrase "Hey whisper," and the text gets enhanced before it lands at your cursor. The pass adds the commas and periods, breaks the run-on into sentences, and strips the "ums" and false starts you didn't mean to keep. On a local model it runs through Ollama; in cloud mode it's gpt-5-mini by default. You didn't say a single punctuation command — you just talked.

The honest limit is that the AI is guessing your intent. It's good at it — for ordinary prose I'd trust it nine times out of ten — but it's making a judgment call about where your sentence ends and where the comma belongs. That's the right trade for an email or a draft, where the meaning survives a comma in a slightly different spot. It's the wrong trade when the comma's position changes the meaning, which is the entire next section.

The same speak-then-clean flow pays off well beyond punctuation — you can also type faster with voice across every app so a paragraph you'd have typed becomes a few spoken sentences that arrive already cleaned up.

When to say the punctuation yourself

Sometimes the right move is the manual command, and pretending the AI handles everything would be dishonest. The rule of thumb: when the exact position of a mark carries meaning, say it yourself. The AI guesses well, but it guesses — and there are places you don't want a guess.

Three cases where I name the marks out loud. First, anything where punctuation changes the meaning — a contract clause, a medical instruction, the difference between "let's eat, Grandma" and the version where Grandma is dinner. Second, code and structured text, where a comma or a colon is syntax, not style, and the AI has no idea you're writing a config file. Third, specific symbols the auto-punctuation won't reach for on its own — a colon before a list, an exclamation point you actually want, a question mark on a sentence that's grammatically a statement. In those spots, the spoken command is faster and surer than dictating, reading the result, and fixing it.

And for short stuff, the built-ins are plenty. If you're firing off a two-line text, Windows Voice Typing (Windows key + H) and macOS Dictation both add basic punctuation on their own, and you don't need to install anything. The catch with Windows Voice Typing is that it routes through Microsoft's servers and needs an internet connection, so it isn't an offline option. Reach for a dedicated tool when the writing gets long, multilingual, or private — or when you simply never want to say the word "period" out loud again.

The same trade-off shows up on a Mac, where Dictation's built-in punctuation covers short notes — the logic in voice to text on a Mac is the same: name the marks when placement matters, let the tool handle it when it doesn't.

So that's the whole thing. Say "comma," "period," "question mark," "new line," "new paragraph" when you want the mark in an exact spot — and let an AI pass do it for you when you just want to talk and have it read like English. I dictated most of this guide the second way, said "Hey whisper" instead of narrating four hundred commas, and only reached for the manual commands when I was writing out the example sentences — the one place in this guide where saying "period" out loud was, finally, the point.