You speak a long, complete thought into your Mac, and what comes back is a single wall of lowercase text: no commas, no capital letters, no paragraph breaks. To fix it you either narrate every “comma, period, new paragraph” out loud, which breaks your flow, or you clean it up by hand afterward. Neither is what you signed up for. Here is how dictation can format a long block for you, automatically, and what actually happens under the hood so you know exactly what to expect.
Two layers do two different jobs
It helps to know that clean output comes from two separate steps, because they behave differently.
The first layer handles numbers, dates, times, currencies, percentages, ordinals and phone numbers. Say “twelve thirty pm” and you get 12:30 pm; say “may first twenty twenty six” and you get May 1, 2026. This runs with zero setup, no toggle to find, and it runs last so nothing re-garbles it. It is also predictable: the same spoken input always gives the same tidy result. One honest caveat: this automatic number and date formatting works in English and French only today.
The second layer is the one that solves the wall-of-text problem: commas, full punctuation, capitalization, spelling fixes and filler-word removal. This comes from on-device AI, and it only kicks in once you turn it on.
Turn on the layer that punctuates and paragraphs
The punctuation and structure layer lives behind a setting called Auto-correct, which is off by default. Its description is plain: it fixes spelling, grammar, and removes filler words. Flip it on and your long dictations come back with sentences that are capitalized and punctuated for you, so you can stop narrating spoken punctuation commands and stop chasing stray lowercase words to fix dictation capitalization by hand.
For a long block, it also groups sentences into paragraphs by topic. When the idea shifts, a new paragraph begins. This is meaning-based, not a reaction to how long you paused, so a thoughtful pause mid-sentence will not chop your text in two. Short remarks are left alone as one tidy sentence on purpose. Paragraph structuring needs a longer input, roughly eight words or more, before it starts organizing.
Keep it on everywhere, or only where you need it
Auto-correct is a global switch, but you do not have to treat every app the same. With per-app dictation profiles you can set behavior per application: light clean-up in one place, a more polished rewrite or on-device translation in another. Turning on a per-app rule is an alternative to the global toggle, so pick whichever fits how you work rather than layering both.
Nothing leaves your Mac
All of this runs locally. Your voice and the resulting text never leave the device, which matters when you are dictating drafts, notes, or anything you would not paste into a cloud service. Dictato works this way by default: it uses its own transcription engine, formats on-device, and is pay-once lifetime with no account to create. You speak the long thought once, and you get back something you can actually read.