How accurate is Apple Dictation, really?

What Apple Dictation gets genuinely right

Let's start with credit where it is due, because a lot of "Apple Dictation is bad" content is written by people selling you something. It is free and already installed on every iPhone and Mac — zero setup, zero accounts. On recent iPhones and Apple silicon Macs, dictation for supported languages runs on-device, which is what makes it fast and lets it keep working without a connection. Length itself is not a constraint either way — dictation just keeps going until it detects 30 seconds of silence. In supported languages it also punctuates for you, inserting commas, periods and question marks as you speak.

On iPhone, the keyboard even stays open while you dictate, so you can fix a word by touch without breaking your flow. For a text message, a search box or a quick note, Apple Dictation is close to effortless — and it would be dishonest to pretend otherwise.

Weakness 1: it types what you said, not what you meant

Apple Dictation is a transcription engine, not a writing tool. Speak naturally for thirty seconds and your draft contains the "umms", the false starts, the mid-sentence corrections and the repeated words — because you actually said them. Nothing rewrites the transcript into the message you intended. This is the single biggest reason dictated drafts feel unusable for anything longer than a text: the recognition may be fine while the writing is not.

Weakness 2: punctuation weakens as the text gets longer

Automatic punctuation is genuinely useful on short sentences. On longer passages it gets inconsistent: commas land in odd places, periods go missing, and you get run-on paragraphs that take real editing time to untangle. Anything beyond the basic marks — colons, dashes, quotes, "new paragraph" — you still have to speak out loud, which is exactly the kind of cognitive overhead that makes people give up on dictation.

Weakness 3: proper nouns, brands and jargon

Names of people, companies, products, medical or legal terms — this is where Apple Dictation misses most consistently, and there is no custom-vocabulary feature to teach it your world. If your writing is full of "Kubernetes", "Lisbeth" or "SaaS churn", expect to correct those by hand every single time. Whatever tool you dictate with, names, numbers and dates are the three things worth proofreading before you hit send.

Weakness 4: the environment matters more than the engine

Background noise, crosstalk, speaking far from the mic, or a mediocre Bluetooth headset will degrade any speech engine, Apple's included. If dictation seems suddenly worse, check the boring things first: the room, the mic, and whether the right dictation language is selected. A surprising share of "dictation got worse" complaints are really "I started dictating in the car".

A two-minute test you can run yourself

Skip the benchmarks and measure the thing you actually care about: your corrections. First, dictate a normal text-message reply. Then dictate a three-paragraph email you would really send — context, an ask, a deadline. Count the edits each one needs.

• If the message was clean and the email needed a handful of fixes, the built-in dictation is enough for how you use it. Keep it.
• If the email came out as a wall of run-ons you had to rewrite, you have found the boundary of raw speech-to-text — no amount of speaking "more carefully" moves it.

The iOS 27 beta: Apple is closing part of the gap

Apple knows all of this. The iOS 27 and macOS 27 betas include an "Advanced AI Dictation" built on a new on-device model, with more reliable punctuation and capitalization — and it strips out filler words and false starts. In Apple's own published side-by-side evaluations, the new model was preferred over the shipping system by a wide margin (44.7% to 17.6% on overall quality). Early hands-on reports confirm the filler-word cleanup works.

The caveats matter, though. As of the current betas it is off by default and limited to recent hardware — think iPhone 17 Pro and iPhone Air, or Macs with an M3 or later and 12 GB of RAM. And it remains a better transcriber: there is still no tone control, no reformatting toward what you meant to write, and no translation. Cleaner literal text, same category of tool.

When built-in is enough — and when it isn't

If you dictate a few messages a day, keep Apple Dictation. It is free, private on recent hardware, and genuinely good at short bursts. A dedicated app earns its place when dictation becomes how you write — emails, documents, messages in more than one language — because what you need then is not a slightly better transcript, it is a finished draft.

That is the category Swifly is in. Swifly is an AI voice dictation app for Mac, Windows, iPhone and Android that turns speech into polished, correctly formatted text in any app — with instant translation. You talk naturally, filler words and all; an AI pass turns it into clean text in the tone you pick — formal, casual, very casual or excited — and one click translates what you just said. To be equally honest about our own trade-off: Swifly processes audio in the cloud, so if fully offline dictation is a hard requirement, Apple's on-device engine — or a local-model app from our Mac dictation comparison — is the better fit.