Sorted by one question: does it type what you said?

Most round-ups in this category sort by accuracy or by price. This one sorts by a single axis that almost nobody publishes: whether the app gives you the words you spoke, or a cleaner version of them. It matters because the difference is invisible in normal use — an edited sentence is well-formed and plausible, so you have no reason to look twice — and because it is usually a default rather than a choice.

Everything below was checked against each vendor's own site on 10 September 2026, and every claim links to the page it came from. Prices and feature sets in this category change often; re-check before buying.

The three groups

Verbatim by default. Apple Dictation, Handy, Coii VoiceInput.

Editing is a feature you may be able to switch off. MacWhisper, VoiceInk.

Editing is the product. Wispr Flow, Aqua Voice, superwhisper.

None of those is a ranking. The third group contains the most capable tools here by almost every other measure, and if you want a polished draft rather than a faithful one, that is where to buy.

Apple Dictation — free, already installed

The one everybody has. It transcribes what it hears without an editing pass: no cleanup, no tone adjustment, no restructuring. For someone who needs verbatim capture a few times a week it may be the entire answer, and any round-up that skips past it in the first paragraph is selling something.

Where it stops being enough is not accuracy. It is what happens around the transcription — which app receives the text, what happens on a long dictation, and how it behaves when you are moving between windows. Apple Dictation against a paid app works through where that line falls, and for a lot of people it never does.

Pick it if verbatim capture is occasional and you would rather not install anything.

Handy — free, open source

Handy is a free and open-source speech-to-text tool: press a shortcut, speak, and the transcription is pasted into the focused field. Its site describes transcribing and pasting, with the voice data staying on your computer, and does not advertise any editing or cleanup pass.

It is the most direct answer to "I want verbatim text and I do not want to pay". If that is the requirement in full, look at it before anything on this page with a price attached, including ours.

Pick it if free and verbatim are the two requirements, and you are comfortable with an open-source tool.

Coii VoiceInput — $19 once

Filler removal ships off, and the post-processing is a fixed chain of text filters rather than a model: repeat collapsing, your own dictionary, optional filler removal, punctuation and spacing. Nothing in that chain turns one phrase into a different phrase, because nothing in it knows what a phrase means. Both the raw transcript and the processed one are stored on every history row, so a filter can be wrong without costing you the words.

The transcript is written to history before the engine is asked anything else, which means a slow model, a crashed engine or a paste that never lands cannot lose a sentence. Nothing is uploaded — no audio, no transcripts — and it works with the network off. macOS 13 or later, three devices, 30-day trial, no account.

What it does not do: it has no vocabulary that learns, no meeting notes, no voice commands, no Windows, iOS or Android, and no AI edit pass at all. If you wanted a clean formatted draft, this is the wrong purchase and the tools in the third group below are the right one.

Pick it if you want your phrasing intact, nothing leaving the machine, and one payment.

MacWhisper — free tier, €64 once for Pro

A local Mac app with a one-time Pro licence, which makes it the closest thing here to this app on the commercial side. Its site lists auto cleanup and automatic grammar improvement among Pro features, along with removal of filler words, so the default posture is toward tidying rather than verbatim — check in the app whether each of those can be switched off before assuming it fits this list.

It does a good deal this app does not, including transcribing existing audio files, which is a different job from typing with your voice. The full comparison covers where the two diverge.

Pick it if you also transcribe recordings and want a one-time licence, and you have confirmed the cleanup settings suit you.

VoiceInk — $25 to $49 once, open source

VoiceInk is open source, macOS only, Apple Silicon, and sold as a one-time purchase across three device tiers. On the verbatim axis it sits in the middle: its own feature list includes removing filler words and repetition, restructuring rambling notes, and rewriting selected text on a voice command, so out of the box it is an editing tool rather than a faithful one.

Being open source, it is also the one where you can read exactly what the processing does rather than inferring it, which is worth something on a page about invisible edits. The comparison covers the rest.

Pick it if you want an auditable codebase and are happy with the editing features, or intend to configure them away.

Wispr Flow — free tier, $15/month Pro, $23/month Growth

The most capable tool in this round-up and the clearest example of the third group. It advertises automatic removal of filler words, added punctuation, formatting as you speak, catching mid-sentence self-corrections, style settings, and a vocabulary that learns your names and jargon as you go. It runs on Mac, Windows, iPhone and Android, claims 100+ languages, and publishes SOC 2 Type II, ISO 27001 and HIPAA certification.

Every one of those is a reason to buy it. It is on this page because the tidying is the design rather than a setting, which is exactly right if you want a clean draft and exactly wrong if your phrasing is the content. The head-to-head is the page that works through both sides properly.

Pick it if you dictate on more than a Mac, or you want polished output without editing it yourself.

Aqua Voice — free tier, $8/month Pro, $24/month Max

Aqua Voice advertises refining text in real time, adapting writing style to the destination, and formatting to suit the medium, with a realtime mode on the higher tier. Mac, Windows and iOS.

It is the purest expression of the editing approach on this list: the pitch is explicitly that what lands is better than what you said. If that is what you want, it is the one built entirely around it.

Pick it if real-time polish is the feature you are shopping for.

superwhisper — free tier, paid tiers including a lifetime option

A local-first Mac app and the closest competitor here in philosophy — it runs on your machine, which means "nothing is uploaded" is not a difference between us. Its site describes predefined modes that optimise tone, structure and formatting, along with email, message and custom modes, so on the verbatim axis it belongs in the third group despite running locally.

Its pricing is presented on the home page rather than a separate page and is easy to misread; check it there directly rather than trusting a figure quoted elsewhere, including here.

Pick it if you want local processing with per-context formatting modes.

Why this list is short

Eight tools, and only three in the first group. That is not a sampling accident — it is where the category has gone.

Transcription accuracy has become close to a solved problem for clear speech into a decent microphone, and once the engines stop being the differentiator, the thing left to sell is what happens to the words afterwards. Tidying is demonstrable in a thirty-second product video in a way that a half-percent accuracy difference is not. So every new entrant ships an editing pass, and the tools that merely type what you said read, in a feature comparison, as the ones missing features.

That is worth knowing when you shop, because it means the axis this page sorts on is one almost no comparison table includes. You will find rows for languages supported, platforms, price and offline use. You will rarely find a row for whether the app changes your phrasing, and when the answer is yes it is described as polish rather than as a change.

The practical upshot: the question has to be asked directly of each tool, and the answer is in the feature list rather than the specifications table. Look for cleanup, formatting, tone, style, grammar, modes, and any promise that what lands reads better than what you said.

The table

Checked 10 September 2026. Verbatim by default means the app does not run an editing pass over your phrasing unless you ask it to.

What you give up by choosing verbatim

A round-up that only lists the upside of its own side is an advertisement, so this section is the cost, and it is not small.

You edit more. An editing tool hands you a clean paragraph; a verbatim one hands you what you said, fillers and dead ends included. Over a hundred emails a week that is a real amount of work moved from the app onto you.

Second-language writing gets harder. If you rely on a tool to fix agreement and article errors as you speak, the verbatim group removes something you were using every day, and none of the three options above replaces it.

No formatting for free. Numbered lists, structured emails and destination-appropriate layouts are things the third group does and this group does not. If your dictation is mostly correspondence, that gap is felt immediately.

Fewer platforms, usually. The verbatim group here is Mac-heavy. Dictation is a habit, and a habit that stops at your phone has a hole in it.

The case for choosing it anyway comes down to a narrow question: is there writing you do where the specific words carry the work? Quotes, prompts, clinical hedging, legal precision, anything with a voice. If yes, no amount of polish compensates for a sentence that now means something slightly different. If no, the third group is genuinely the better buy and this page has been an argument against your own interests.

The two questions that are not on this axis

Whether an app rewrites you is one property. Two others get confused with it and should not be.

Where the audio goes. Local processing and verbatim output are unrelated. superwhisper runs on your machine and reshapes your text; a cloud service could transcribe you faithfully. The privacy question is its own question.

What happens when something fails. A transcript that a tool never wrote down because a model timed out is a different loss from one that was edited. Ask both: does the app keep what it originally heard, and does it keep anything at all when the engine falls over halfway through a sentence?

An app that stores only its own output has answered both questions badly, and you will find out on the day it matters.

How to decide

Run the four-minute test on whatever you use now: dictate a sentence with two hedges and a mid-sentence correction, one with deliberate fillers, and one long run-on with no spoken punctuation. How to dictate verbatim on a Mac has the exact sentences and what each result tells you, and does a dictation app change what you said covers the feature-list vocabulary that gives it away before you install anything.

If the result surprises you, the question is not which app is best. It is whether you want an editor in the loop at all — the longer argument is here, and if you land on "no", the cheapest answer is free and the one-time answers are on this page.