Google's dictation lives inside a document

The honest summary first, because it decides whether the rest of this page is worth your time. Google's voice typing costs nothing, it recognises well over a hundred languages, and it can edit — "select paragraph", "bold", "insert table", "go to end of document" — which this app cannot do in any form. It also exists only inside Google Docs and the speaker notes of Google Slides, in a browser tab, entered from a menu. If most of your writing happens in a document, it may be all you need. If your writing is spread across a mail client, a chat window, an editor and a terminal, it cannot reach any of them.

What Google says it is

Worth quoting from Google's own support page rather than paraphrasing, because the scope is the entire argument. You can use your voice "to type and edit your document in Google Docs and your speaker notes and captions in Google Slides", and the feature "works with the latest versions of: Chrome, Edge, Safari".

You turn it on from Tools → Voice typing, then click the microphone icon when you are ready to speak. You stop it by saying "Stop listening", and you start again by saying "Resume". Their page also notes that your "computer microphone needs to be on and working", which on a Mac means the input has to be selected and permitted before any of this happens at all.

One more line from the same page, and it is the one most reviews skip: "When you turn on voice typing or captions, your web browser controls the speech-to-text service. It determines how your speech is processed and then sends the text to Google Docs or Google Slides." Read that carefully. The service doing the recognising is the browser's, not the document's, and Google is telling you plainly that the processing is out of their hands and out of yours. They document no offline mode for it.

Where Google voice typing is better

Stated properly and first, because a comparison that hides this is an advertisement.

It is free, and there is nothing to install. No download, no licence, no trial to remember to cancel. If you already have a Google account and a supported browser, you have it.

It edits. This is the difference people underrate. Google's command list covers selection ("Select paragraph", "Select all"), formatting ("Bold", "Italicize"), navigation ("Go to end of document", "Move forward five words"), tables ("Insert table") and the clipboard ("Copy", "Paste"). Those are real operations on an existing document, and this app has no equivalent for a single one of them. It writes new text; it has no vocabulary for changing text that is already there.

It speaks a very long list of languages. Google's page lists well over a hundred entries, from Afrikaans to Zulu, with regional variants for Arabic, English, Spanish and others. Punctuation commands, they warn, "might not be available in every language" — but the recognition list itself is longer than anything a small Mac app is going to publish.

It runs wherever the browser runs. A Chromebook, a Windows laptop, a shared machine in a library. This app is macOS 13 or later and nothing else, and there is no argument to make on that row.

Two limits Google states themselves, which are fair to mention because they come from the same page: voice commands "are available only in English", and "the account language and document language must both be English" for them to work; and voice commands are not available in Slides speaker notes at all. So the editing power that is Google's strongest card applies to English documents in Docs.

The table

Five clear rows to Google, and the free one is worth more than the other four put together for a lot of people. Our side of the table is four rows wide: everywhere else on the Mac, no mode, nothing uploaded, and the words written down before anything can go wrong with them.

The other places you write

Here is the test that settles it, and it takes a day rather than an argument. Tomorrow, every time you type something longer than a sentence, note which application you were in. Most people who try this are surprised by two things: how many different applications appear on the list, and how few of the entries are a document.

That last row matters. A Google Doc open in a browser is a text field like any other, so a Mac-wide dictation key fills it in exactly the way it fills a Mail reply. The relationship is not symmetrical: Google's voice typing cannot leave the tab, and there is no setting, extension or workaround on their page that changes that.

If your work genuinely is one long document — a dissertation, a manuscript, a report you live in for three weeks — this section does not apply to you and Google's tool is a very reasonable answer. For everybody else, the tool that only works in one place is the tool you stop reaching for by Thursday.

Menus, modes, and the cost of both

There is a second difference, smaller in a feature list and larger in daily use: how the two are turned on.

Google's is a mode. You open a menu, you click a microphone, you speak, and you say "Stop listening" when you are finished — or you forget to, and the microphone stays live while you talk to somebody at the door. The mode has to be entered before you start and exited after you finish, and both of those are things you have to remember while composing a sentence, which is exactly when you have no attention spare.

Holding a key is not a mode. Press it, talk, let go. There is no state to be in the wrong one of, nothing left running, and no menu between the thought and the text. It sounds like a trivial ergonomic difference and it is the single biggest reason the daily press count on a held-key tool ends up so much higher than the session count on a menu-driven one. A four-second dictation is worth doing when it costs one keypress and not worth doing when it costs two clicks and a spoken command.

What happens to your voice

This is where the two products differ structurally rather than in degree, and where the honest formulation matters.

Google tells you that the browser controls the speech-to-text service and "determines how your speech is processed". That is their sentence, not an inference of ours, and what it means in practice depends on which browser you opened the document in. It is a genuinely unusual arrangement, and it is worth knowing about before you dictate anything sensitive into a tab.

Coii VoiceInput has a shorter answer. The audio is processed on your Mac and nothing is uploaded — no recordings, no transcripts, no text. The model weights are fetched once from a public model host, and after that the network is used for licence activation and a clock check at launch, and for nothing else. Turn the Wi-Fi off and everything still works, on a plane, on a locked-down laptop, in a building where the guest network is the only network and you are not putting a client's paragraph through it.

That is not a claim about accuracy or speed. It is a claim about where the words are, and it is the row where a local app is structurally ahead of anything that runs through a browser — see also Wispr Flow, where the same argument applies to a hosted service, or superwhisper, where it does not apply at all because that app is local too.

When the recognition fails

Google's troubleshooting advice for their own tool is honest and familiar: if you see "We're having trouble hearing you", move to a quiet room, plug in an external microphone, adjust the input volume. Every product in this category has some version of that page, this one included, because half of dictation accuracy is the microphone and not the software.

The more interesting failure is the other one: the transcription that arrived and then went missing. A tab that reloaded, a session that expired, a document that had not synced. What a local tool can do about this is specific.

Release the key and the transcript is written to your own history before the rest of the sequence runs. Everything that might go wrong afterwards — the app you were aiming at quit, the paste landed oddly, the field rejected it — happens after the sentence has already been saved, so the worst case is a copy-and-paste from a list on your Mac rather than a paragraph you have to reconstruct from memory. That history is local, it is searchable, and it is yours whether or not any browser is running.

The second behaviour is the one people notice in week two: the destination is fixed the moment the key goes down. Start talking into a chat window and that is where the text goes, even if a notification steals focus while you are still mid-sentence.

The languages question, honestly

The language list is the row where Google is furthest ahead and it deserves more than a table cell.

Their page lists well over a hundred entries and it goes deep into regional variants — several Arabics, several Englishes, several Spanishes, Urdu split between India and Pakistan. If you write in a language that a small Mac app is unlikely to have prioritised, that list is not a marketing number, it is the difference between a usable tool and an unusable one, and no amount of architecture on our side changes it.

Two caveats from Google's own page keep it in proportion. Punctuation, they warn, "might not be available in every language", so the language being on the list does not guarantee that "comma" and "new paragraph" do anything in it. And the command vocabulary — the selecting, formatting and navigating that is Google's real advantage — is English only, and needs both your account language and your document language to be English. So the hundred-language claim and the edit-by-voice claim apply to overlapping but different sets of users.

The practical advice is unglamorous: whatever you are considering, dictate three paragraphs of your own actual writing in your own actual language and read the result. Both of these tools are free to test for that, one because it is free and one because the trial is thirty days, and three paragraphs will tell you more than any comparison table including this one.

A test that takes one week

If you are genuinely undecided, do not compare features. Compare where you write.

Spend a week with Google's voice typing and use it properly — open the panel, learn four or five of the commands, say "Stop listening" when you finish. At the end of the week, answer two questions. How many times did you use it? And how many times did you want to dictate something and could not, because you were in a mail client, a chat window, a form, a ticket or a terminal?

If the second number is small, you write in documents and you should stop here. Google's tool is free, it edits, it speaks your language, and there is nothing in a $19 purchase that improves on that for your particular day.

If the second number is large — and for most people who are not full-time document writers it is several times the first — then the feature you are missing is not accuracy or vocabulary. It is reach, and reach is the only thing this comparison is really about.

What each one costs

Google's is free. There is no price to argue with and no trial to convert. If free is the requirement, the comparison is over and you should use it, or use Apple Dictation, which is also free, also already installed, and is not confined to a browser tab.

Coii VoiceInput is $19, once. Three Macs on the licence, updates included, nothing renews, and there is no tier above it. The trial is thirty days of the complete app with no card and no account, which is long enough to answer the only question that matters — how many times a day do you actually press it. Start there rather than from a table.

Against a monthly service that arithmetic is the headline. Against a free feature it is not an argument at all, and the reason to pay is the reach and the behaviour above, or nothing.

Using both, which is the common answer

Nothing here is exclusive, and the setup a lot of people land on is the obvious one. Use Google's commands when you are inside a Doc restructuring English prose, because "select paragraph" and "insert table" are genuinely useful and this app will never offer them. Use a held key everywhere else, including inside the same document when what you want is a paragraph rather than an edit.

They do not collide. One is a browser feature reached from a menu; the other is a system-wide keypress. The only thing to watch is the microphone: if Google's voice typing is still listening because you forgot to say "Stop listening", and you then hold the dictation key, two things are recording at once and the results will be confusing. Say it, or close the panel, before you switch.

Which one to use

Google voice typing if your writing is mostly Google Docs, you want to edit and format by voice, you need a language a small app is unlikely to ship, you are not on a Mac, or free is not negotiable. It is a capable feature and it costs nothing.

Coii VoiceInput if you write in six applications a day, you want one key instead of a menu, you need it to work with the network off, or the text you dictate is not something you want a browser's speech service to determine the processing of. macOS 13 or later, $19 once, and thirty days to find out.

If neither is quite it: Apple Dictation is free and system-wide, Murmur is free and open source, dictating prompts rather than documents is a different question, and if you would rather build the thing yourself, a Whisper script is a real option with a real cost attached.