Short answer: Apple Intelligence and dictation are separate systems. Apple Intelligence adds writing tools that rewrite, proofread, and summarize text after it exists. It does not replace the speech-to-text engine, so the familiar dictation complaints about proper nouns, session length, and third-party apps largely remain.
Since Apple Intelligence arrived, a predictable question has followed every conversation about voice typing on Apple devices: does this fix dictation? People who dictate daily have a list of grievances, and it is reasonable to hope that a large platform-level push into on-device intelligence would clear some of them.
The honest answer is that it clears a few and leaves most of them alone, because Apple Intelligence and dictation are not the same system and were never designed to be. Understanding the split explains both why some things got noticeably better and why the specific thing that annoys you probably did not.
Two systems, not one
Dictation on Apple devices is a speech-to-text service. You press the microphone key or the dictation shortcut, it listens, and it converts audio into characters in a text field. Its job ends the moment the words land. It has no opinion about whether the sentence is any good.
Apple Intelligence is a layer that operates on text that already exists. Its writing features let you select a block of text and ask for a proofread, a rewrite in a different tone, or a summary. Everything it does happens after the words are on the page, regardless of how they got there. It works identically whether you typed the paragraph, pasted it, or dictated it.
That architectural boundary is the whole story. When people ask whether Apple Intelligence improves dictation, they usually mean one of these:
- Does it hear my words more accurately?
- Does it get proper nouns and jargon right?
- Does it stop cutting me off mid-thought?
- Does it work in every app?
Those are all questions about the speech-to-text layer. The writing tools sit above it and cannot answer any of them. What the writing tools genuinely improve is the cleanup step afterward, which is real value but a different kind of value than most people are asking about.
What did get better
Three things are worth naming honestly, because dismissing the whole feature set would be as wrong as overselling it.
Cleanup after the fact
Dictated text has characteristic problems. It runs long. It repeats. It preserves verbal scaffolding like "so basically" and "the thing is" that reads fine aloud and badly on a page. Selecting a dictated paragraph and asking for a tightened rewrite handles exactly this class of problem, and it does it in one step instead of five minutes of manual trimming.
This is the strongest pairing between the two systems: dictate for speed, then clean up in place. If you have historically avoided dictation because the raw output embarrassed you, this is a legitimate reason to revisit it.
Tone adjustment
People dictate more casually than they type. When the destination is a work email or a client message, the register is often slightly off. Rewriting for a more formal tone in a single action removes a real friction point, particularly for people who dictate on a phone between other tasks.
One caution here that applies to any automated rewrite: it changes your words. For a routine scheduling email that is fine. For anything where the exact phrasing carries weight, read the result rather than accepting it. We take a firm position on this generally, which is that a dictation tool's job is to transcribe what you said, not to decide what you meant. Polishing should be a separate, deliberate step you invoke, not something that happens silently to your sentences.
Proofreading
Catching the grammatical wreckage that comes from restarting a sentence mid-dictation is genuinely useful, and it is a very common dictation artifact. This is the least controversial of the three because it corrects errors rather than making stylistic choices for you.
What did not change
Proper nouns and specialized vocabulary
This remains the number one complaint about built-in dictation, and it is structurally unfixed by a writing layer. If the engine hears your colleague's surname and produces a common word that sounds similar, a downstream proofread will not catch it. The result is grammatically perfect and semantically wrong, which is the worst possible failure mode because it does not look like an error.
The same applies to industry jargon, product names, medical and legal terms, and anything else that is rare in general text but constant in your particular work. A rewrite tool operating on the output cannot recover a word that was never captured.
The only real fix for this class of error is a vocabulary the system knows about in advance. Voice Keyboard Pro handles it with Smart Vocabulary, a personal dictionary with replacement rules that you build up from your own corrections. It is unglamorous and it works, because it addresses the problem at the point where the error is introduced rather than after.
Session length
If your dictation stops on you before you have finished a thought, that is a property of the dictation session, not of the writing tools. Nothing in a rewrite layer extends how long the microphone stays open. We wrote about this specific frustration in the Mac dictation time limit fix, and the guidance there still holds.
Third-party app coverage
Built-in dictation behaves differently across apps depending on how each one implements text input. Some fields work perfectly, some drop characters, some do not trigger at all. This is an integration surface issue and it is untouched by anything happening at the writing-tools layer. See Mac dictation not working in third-party apps for the practical workarounds.
Hardware requirements
Apple Intelligence requires relatively recent hardware. Broadly, that means recent iPhone Pro models and newer iPhone generations, and Macs and iPads with Apple silicon. If you are on an older Intel Mac or an iPhone a few generations back, none of this applies to you at all, and the dictation experience you have today is the dictation experience you will keep having.
This is worth checking before you form an opinion, because a surprising number of "Apple Intelligence did nothing for my dictation" complaints come from devices where the feature set is not available in the first place.
Apple Intelligence is not Siri, and neither is dictation
Three things get conflated constantly, so it is worth separating them cleanly.
- Siri takes voice commands and performs actions. You speak an instruction and something happens.
- Dictation takes voice and produces text. You speak content and it appears where your cursor is.
- Apple Intelligence writing tools take existing text and transform it. No voice involved.
People frequently blame dictation for a Siri limitation or credit Apple Intelligence for a dictation behavior. If you are trying to diagnose why voice input is not doing what you want, figuring out which of these three you are actually using solves the problem more often than any setting change. We broke this down further in Siri vs dictation on iPhone.
Where a dedicated dictation app still fits
The case for a separate dictation tool is not that Apple's approach is bad. It is that the two systems optimize for different things. Apple ships a general-purpose capability that must work acceptably for everyone on the platform. A dedicated tool can make choices that would be wrong as a default but are right for people who dictate as their primary input method.
Concretely, those choices look like this.
Hold-to-talk instead of toggle. Holding a hotkey while you speak and releasing when you are done maps to how people actually dictate: in bursts, between thoughts. There is no session to start, no session to remember to end, and no ambiguity about whether the microphone is still live. On a Mac, Voice Keyboard Pro works this way system-wide, so the same gesture produces text at the cursor in any application, whether that is a native app, a browser field, or a terminal.
A vocabulary that is actually yours. Smart Vocabulary lets you register the names, terms, and abbreviations you use and define replacement rules for them. This is the difference between correcting the same word two hundred times a month and correcting it twice.
Meeting Mode. Speaker detection and AI notes for meetings, with calendar meeting detection so a recurring call does not require you to remember to start anything. This is a different job from dictation and it is not something the built-in text-input path attempts.
Keyboard-level dictation on iPhone. The iPhone side is a custom keyboard with a built-in microphone button, which means it works in any iOS app that accepts keyboard input, including apps whose own text handling is idiosyncratic. It also carries Voice Edit, where you speak the change you want rather than tapping to fix text, and two-way translation across 24 languages while you dictate. Swipe typing is there for the moments when speaking is not an option.
A clear privacy boundary. Our servers store only operational pings. No audio and no transcript content is retained. That is a design decision rather than a marketing line, and it is the reason we can state it plainly.
How to test whether it changed anything for you
Opinions about this are unusually loud and unusually unmeasured. If you want to know what actually improved on your own device, with your own voice and your own words, the test takes about ten minutes and it is worth more than any general write-up including this one.
Pick three paragraphs of real work. Not a sample sentence and not a passage from a book. Something you would genuinely send: an email to a client with their actual name in it, a set of notes containing the terminology you use every day, a message with the product names or place names that recur in your job. The whole point is to include the vocabulary that general-purpose systems have never seen.
Dictate all three, twice, without editing as you go. Then count two separate numbers.
- Correction count. How many individual fixes does the raw text need before you would send it? Weight proper nouns heavily, because those are the errors that survive proofreading and reach the recipient.
- Interruption count. How many times did you have to restart, retrigger, or work around the tool rather than just speak? These are the errors that determine whether you are still dictating in a month, and they never show up in accuracy percentages.
Most people find the first number is dominated by a small, stable set of words that recur constantly, which is exactly the set a personal dictionary eliminates. The second number is usually about session behavior and app coverage rather than accuracy at all. Knowing which of the two is your actual problem tells you whether you need a better capture tool or just a better cleanup habit.
One methodological note: do not run the test in a quiet room if you do not work in one. Acoustics change results substantially, and a result you gathered in ideal conditions will not predict your Tuesday afternoon.
So should you use both?
Yes, and the combination is better than either alone. They are complementary rather than competing, once you stop expecting one to do the other's job.
A workable division of labor:
- Capture with a dedicated dictation tool. This is where accuracy, vocabulary, speed, and app coverage matter, and where the differences between tools are largest.
- Polish with whatever writing assistance you have available. Proofread and tone adjustment are genuinely useful on dictated text, and invoking them deliberately on a finished draft is the right shape for that feature.
- Keep them separate. Resist tools that silently rewrite while transcribing. When capture and editing are fused, you lose the ability to tell whether a sentence is what you said or what something decided you meant.
The honest summary
Apple Intelligence is a meaningful addition to writing on Apple devices, and if you dictate, the cleanup features will save you time. But it did not rebuild dictation, and the specific things that make people abandon voice typing, which are almost always accuracy on the words that matter most to them and reliability across the apps they actually use, sit below the layer it operates on.
If your complaint is "my dictated text reads badly," a writing tool helps. If your complaint is "it gets my clients' names wrong," or "it stops before I finish," or "it does not work in the app I live in," you need something that operates at the capture layer.
A rewrite tool cannot recover a word that was never captured correctly. It can only make the wrong word into a better sentence.
If you want to test the capture layer against your own vocabulary, Voice Keyboard Pro has a free tier with daily limits, and Pro is $4.99 per month or $34.99 per year. A useful comparison is to dictate the same three paragraphs of your real work, with real names and real terminology, through both approaches and count the corrections. That number is the whole decision. For more on the built-in baseline, see Apple dictation alternatives and voice keyboard vs Apple dictation on iPhone.