Dictating emails on a Mac: what works and what breaks

There is no such thing as email dictation on a Mac. There is one system-wide feature, and your mail app inherits it. Open the Apple menu, choose System Settings, click Keyboard in the sidebar, then turn Dictation on. macOS may download a language file the first time. The full walkthrough, including the settings people miss, is in the guide to how to dictate on a Mac, so here are only the three things that behave differently inside a mail client.

1. Click into the compose field before you press anything. This is the one that catches everybody. A mail window has a lot of places for focus to sit: the mailbox list, the message list, the reading pane, the search box. If the cursor is not in the body of a message you are writing, there is nowhere for your words to go, and the feature looks broken when it is working perfectly.
2. Know all three ways to start. Press the Microphone key if your keyboard has one, press your dictation shortcut, or use the menu bar: Edit, then Start Dictation. The menu-bar route is worth remembering in Mail because it is the one that cannot be swallowed by an app shortcut.
3. Know the three ways to stop. Press the shortcut again, press Escape, or click the microphone indicator that appears near the cursor. Escape is the one you want mid-email, because it leaves the text you have already dictated in place.

On the shortcut itself: on a MacBook Pro or MacBook Air from 2021 onward, and on the Magic Keyboard, the dictation key is F5 with a microphone icon on it. Apple's Magic Keyboard guide describes pressing the Microphone key to activate dictation, and says you can dictate text wherever you can type, naming Mail among the examples. Without that key, the default is pressing Fn twice. You can change it in System Settings, Keyboard, Dictation, from the Shortcut menu, and the options are walked through in the piece on the Mac dictation keyboard shortcut.

Three documented behaviors matter more for email than for anything else you dictate, and at least one of them is widely reported wrong.

There is no length limit, but there is a silence limit. Apple's support page for dictating messages and documents on Mac says you can dictate text of any length without a timeout, and that Dictation stops automatically when no speech is detected for 30 seconds. Plenty of articles still quote a hard cap of a minute or less, carried over from an older macOS. The cap is not the problem. The silence rule is, and email is where it hurts, because composing an email is mostly not talking. You scroll up to reread what they asked. You check how they spell their surname. You stare at the ceiling deciding how blunt to be. Thirty seconds of that and the session has closed behind you.

Auto-punctuation covers three marks. In supported languages, Dictation inserts commas, periods and question marks for you as you speak. That is the whole list. A colon before a bulleted list, parentheses around an aside, a semicolon, a dash: those still need to be spoken by name. The command list itself lives in the guide to dictating punctuation on a Mac, so it is not repeated here.

On Apple silicon you can type while it listens. Apple states that on a Mac with Apple silicon you can still use the keyboard while you speak, with no need to stop dictation. In email that is the single most useful thing in the whole feature set: when it mangles a client's name, you fix the name with the keyboard and carry on speaking, instead of ending the session, correcting, and starting again.

Outlook is its own story, because the Dictate button on the ribbon sends your audio to Microsoft's cloud speech service and needs a Microsoft 365 sign-in, which is a different trade from anything Apple Dictation makes. That comparison, and the Outlook-specific workflow, is set out in dictating into Outlook on Mac, so this page leaves it there rather than covering it twice.

People searching for how to dictate Gmail on a Mac are usually looking for a button that does not exist, and the results they get do not tell them so. There are two separate things with similar names.

Google's voice typing is a web feature built into Google's editors. Google's own documentation for typing with your voice scopes it to Google Docs and to the speaker notes and captions in Google Slides. Gmail is not on that list. It also runs through the browser, so it needs a supported browser and the site's microphone permission, and its spoken command set is short: period, comma, exclamation point, question mark, new line, new paragraph, with commands available only in English.

macOS Dictation is a system service. To it, a Gmail compose box is just a text field on screen, exactly like a Notes window. That is why it works in Gmail when Google's own feature does not, why it does not care about Chrome's per-site microphone permission, and why it behaves the same in Safari as in Chrome. The one thing to watch is shortcut collision: Gmail is full of single-key shortcuts, so if you have set a custom dictation combination, check it does not already mean something to Gmail.

The practical upshot: if someone tells you to dictate in Google Docs and paste the result into Gmail, that advice only made sense before you knew you could dictate into the compose box directly. The Docs-specific detail, if you do write there, is in dictating into Google Docs.

Dictation demos are always the same sentence: a tidy line of plain prose, spoken clearly, in a quiet room. Real email is not that. It is names, addresses, numbers and structure wrapped around a small amount of prose, which is why people try dictating email, love it for a day, and drift back to typing. These are the seven friction points, and the fix for each.

1. Names and proper nouns. A speech model guesses from sound, and a surname it has never seen is the hardest guess in the language. Aoife, Siobhan, Xiaoyu, Ng, Krzysztof, and every company name that is a made-up word all come out as something plausible and wrong. Getting a client's name wrong in the greeting is worse than a typo anywhere else in the message, because it is the first thing they read. The fix on Apple silicon is the keyboard-while-dictating behavior above: type the greeting, then dictate from the first real sentence onward.
2. Email addresses and URLs read aloud. There is no address-aware handling in either Apple Dictation or Google's voice typing. Apple's list of dictation commands gives you the pieces (at sign, dot, hyphen, underscore, forward slash) but you assemble them by hand, and the result usually arrives with spaces around every symbol and a capital letter at the front because the sentence-casing put one there. A three-part address takes longer to say and then repair than it takes to type. Type addresses and links. Always.
3. Punctuation and paragraph breaks. The three auto-punctuated marks get you through plain sentences. Structure does not come free: a paragraph break needs you to say "new paragraph", a line break needs "new line", and a colon before a list needs "colon". An email that is a wall of text reads as careless no matter how good the sentences are, so the structure is not optional. This is the friction that makes people say dictation is slower than typing, and it is real.
4. Formal sign-offs. Saying "kind regards comma new line" and then your own name out loud is not something anyone enjoys, and your mail app already puts a signature in for you. The sensible answer is to stop dictating at the last real sentence and let the signature do its job. If you sign off by hand instead, make it a text replacement rather than something you speak.
5. Numbers and figures. Speech-to-text has to decide whether you meant a numeral or a word, a date or a list, a decimal point or a period. Invoice totals, reference numbers, times, percentages and version numbers are all coin flips, and in email they are usually the load-bearing part of the message. Say them, then reread them before you send. If the number is the whole point of the email, type it.
6. Mixed-language email. Apple Dictation runs one language at a time. You can add more, and you can switch mid-dictation by clicking the language shown next to the cursor, but that is a manual act per switch. If you write to a German colleague and a French supplier in the same fifteen minutes, you are clicking a language picker between messages. The wider version of this problem is in the guide to multilingual dictation on Mac.
7. Long replies versus one-liners. This is the one people get backwards. Short replies are where dictation feels weakest, because the overhead (click into the field, trigger, speak, stop, fix) is a large fraction of a six-word answer. Long replies are where it pays, because the fixed cost is the same and the prose is the bulk of the work. But long replies are also where you pause to think, and 30 seconds of thinking closes the session. So the built-in feature is best at exactly the length of email it is least likely to survive.

Every one of the seven has the same shape. Dictation is good at prose and bad at data. So stop trying to dictate a whole email and dictate the part that is writing.

• Type: the recipient, the subject line, the greeting with their name in it, any email address, any link, any figure that matters, and the sign-off.
• Speak: everything between the greeting and the sign-off. The explanation, the reasoning, the apology, the context, the answer to the actual question. That is the part that takes you three minutes to type and forty seconds to say.
• Then reread once before sending. Not a full edit. A scan for the four things that go wrong: a name, a number, a missing paragraph break, and a sentence that ran on because you did not pause where you thought you did.

A worked example. The typed part is the frame: "Hi Priya," at the top, your signature at the bottom, the reference number pasted from the invoice. The spoken part is the middle, in one continuous pass: "Thanks for chasing this, and apologies for the delay. I have checked with the team and the delay came from our side, not the courier's. New paragraph. I have credited the shipping cost back to the card on file, which should land in two to three business days. If it is not there by Friday, reply to this message and I will call the payment provider myself. New paragraph. On the wider order, the remaining items are still on schedule for the end of the month and nothing about this changes that."

That is around twenty-five seconds spoken against two or three minutes typed, and the only things left to check are the name at the top, the number pasted in, and whether the two paragraph breaks landed.

Start with focus, because in email that is nearly always it: the cursor is in the message list, the reading pane or the search field, not in the body you are writing. Click into the compose field and try again.

If that is not it, Apple's page on keyboard dictation not working as expected lists the rest, and the order is worth following: confirm Dictation is switched on in Keyboard settings; confirm you are pressing the shortcut you think you are; confirm the right microphone is selected in Sound or Keyboard settings and its input volume is high enough; confirm nothing is physically covering the microphone; and confirm the language and region match what you are actually speaking. Apple also notes that a noisy room or a room with a lot of echo is a real cause, and that a headset microphone helps.

Two more that look like faults and are not. A session that ends on its own after you stop talking is the documented 30-second silence rule, not a crash. And a session that ends when you switch windows is expected: dictation follows the cursor, so clicking away from the compose field takes the target with it. The longer troubleshooting list, including the cases that are genuinely broken, is in the guide to dictation not working on Mac, and the cut-off behavior has its own page on Apple Dictation cutting off.

My bias is on the table: I build Parakeety, a paid Mac dictation app, so I am one of the alternatives to the free feature this whole page has been explaining. That is exactly why the explanation is specific rather than dismissive. Every behavior I have attributed to Apple Dictation above is linked to Apple's own documentation on the page where I claim it, including the one most articles get wrong, and I would rather correct the stale timeout figure than repeat it because it flatters my product.

Parakeety is push-to-talk. You hold the section key, talk, release, and the finished text pastes at the cursor in whatever you were typing into, including a Mail compose window and a Gmail tab. Two of the seven friction points above change shape because of that design. The session lasts exactly as long as you hold the key, so a thinking pause in the middle of a long reply cannot end it. And punctuation comes from the model rather than from spoken keywords, so paragraphs and commas arrive without your saying "new paragraph", which is the difference between talking and reading punctuation aloud. Whether that is worth $25 depends on how much of your day is the inbox. The head-to-head is in Parakeety against Apple Dictation, and the trade-off between holding a key and leaving a microphone open is in push-to-talk versus always-on dictation.

The other five friction points do not go away. Names, addresses, figures, sign-offs and mixed languages are hard for every speech model, mine included. Speak the prose and type the data holds whichever tool you use.

Start with the built-in feature, because it is free and already on your Mac: System Settings, Keyboard, Dictation, then click into a compose window and talk. If what sends you back to typing is the session closing while you think, or saying "new paragraph" out loud twenty times a morning, that is what Parakeety is for. Hold the section key, say the middle of the email, release; it pastes at the cursor already punctuated, in Mail, Outlook, a Gmail tab or anything else. Transcription happens on the Apple Neural Engine, so the contents of your email never leave the machine. It needs Apple silicon and macOS 14 or later. Free for 7 days with no card, then $25 once.

Try Parakeety free →