Short answer: Speak in complete phrases at your normal conversational volume and pace, keep the microphone within arm's length, and finish your sentences instead of trailing off. Do not over-articulate, do not slow down, and do not raise your voice. Clarity comes from context and consistency, not effort.
Almost everyone approaches dictation the way they approach a bad phone line. They slow down. They separate. They. Say. Each. Word. They get louder. And accuracy gets worse, which confirms their suspicion that dictation does not work for them, and they go back to typing.
The instinct is understandable and completely backwards. Speaking clearly for dictation looks much more like speaking normally than most people expect, and the handful of adjustments that genuinely help are not the ones people try first.
Here is what actually moves accuracy, in rough order of how much difference each one makes.
Why over-articulating makes things worse
Start with the mechanism, because it explains six of the seven rules below.
Transcription does not identify words one at a time in isolation. It works over a stretch of speech, and each word's identity is constrained by the words around it. "Their" and "there" sound identical. What separates them is the rest of the sentence. Same with "to," "too," and "two." Same with almost every ambiguous sound in English.
When you say one. Word. At. A. Time, you delete exactly the information that resolves those ambiguities. Each word arrives as an island with no neighbours to constrain it, and the engine falls back on which word is more common in general, which is frequently not the one you meant.
Over-articulation has a second problem. Natural speech has predictable connections between sounds. When you exaggerate every consonant and put a hard boundary around every syllable, you produce a signal that is less like the speech patterns transcription is tuned for, not more. You sound clearer to a human listener and less familiar to the system.
Dictation is not a bad phone line. It is a listener that understands sentences and struggles with fragments.
Rule 1: Speak in phrases, not words
This is the single highest-value change, and it costs nothing.
Say a whole clause before you pause. "I'll send the revised contract over on Thursday morning" as one unit, not seven separate deliveries. If you need to think, think before you start the phrase rather than inside it.
Two things improve at once. Word accuracy goes up, because context is available. And punctuation improves, because automatic punctuation infers clause boundaries partly from where you pause. If you pause randomly mid-sentence, you get commas in strange places, which is the root cause behind dictation inserting random commas.
The practical version: think one phrase ahead. While you are saying the current clause, your brain should be assembling the next one. This is a learnable skill and it takes about a week of regular use. It is also the exact skill that makes people fast at dictation rather than merely functional at it.
Rule 2: Use your normal speaking volume
Raising your voice does not add clarity. It changes your vowel quality, flattens your intonation, compresses your dynamic range, and on many microphones it introduces distortion that makes the signal measurably worse.
If you feel the need to be louder, the real problem is distance or the input device. Fix that instead. A quiet, close voice beats a loud, distant one every time.
There is a health argument here too. Projecting for an hour a day, every day, is how people end up with a tired voice from a tool that was supposed to reduce strain. We covered this in voice strain from dictation, and the summary is that every strain behaviour people develop is downstream of misrecognition. Fix the accuracy and the shouting stops on its own.
Rule 3: Get closer, which beats buying anything
Sound intensity falls off sharply with distance. Halving the gap between your mouth and the microphone makes a much bigger difference than most hardware upgrades, and it is free.
Practically: your laptop's built-in microphone at normal seated distance is fine for a quiet room. Leaning back in a chair three feet away, with the lid angled away from you, is not. Wired earbuds with an inline mic sit much closer to your mouth than any laptop and cost almost nothing.
Consistency matters as much as distance. Moving around while you speak, turning your head to look at a second monitor, gesturing with the hand holding the phone: all of these change the signal mid-sentence. Pick a position and hold it for the duration of the utterance.
Our full breakdown in the best microphone for dictation makes the same point at length, including why four of the top five accuracy improvements cost nothing.
Rule 4: Finish your sentences
Most people trail off. Volume drops through the last three or four words of a sentence, especially when the ending is predictable and the speaker has already mentally moved on.
Those trailing words are the ones that come back wrong or missing. Keep your energy level roughly constant to the final word, then stop. Not louder at the end. Just not quieter.
The related habit is not letting the last word blur into your spoken punctuation. Say the sentence, small beat, say "period." Without the beat, "period" occasionally attaches to the previous word and you get a word you did not say.
Rule 5: Keep your normal pace
Slowing down feels like it should help. It does not, for the same reason as over-articulating: it stretches the natural rhythm that carries information about where words begin and end.
Speak at conversational speed. If you are naturally a fast talker and you are seeing errors, do not slow to a crawl. Come down slightly and, more importantly, put clean boundaries between phrases rather than trying to spread the words out evenly.
The opposite case is worth naming too. If you are reading text aloud rather than composing, most people speed up and flatten out, because reading takes no cognitive effort. Read at the pace you would speak.
Rule 6: Speak your punctuation, and be consistent about it
Automatic punctuation is good and improving, but it is inferring. It will not know that a sentence was meant to be a question if your intonation did not rise, and it cannot know where you wanted a paragraph break.
Say "period," "comma," "question mark," "new paragraph." It takes about a day to stop feeling strange. Once it is automatic, you produce clean text on the first pass, which is the difference between dictation that saves time and dictation that moves the work from typing to fixing.
We have a full command reference in Mac dictation commands, and the paragraph-level commands in dictating new lines and paragraphs. One warning worth repeating: in chat apps, a spoken line break can send the message.
Rule 7: Do not neutralise your accent
This one gets its own rule because the advice people give each other is actively harmful.
Trying to perform a different accent than your own produces inconsistent, unnatural speech, which is worse than any accent spoken naturally. Modern transcription handles a wide range of accents well. What it handles badly is a speaker changing their vowels halfway through a sentence because they are self-conscious.
What does help is making sure the language variant setting matches how you actually speak rather than where you happen to live. A mismatch here produces a specific and recognisable failure: the words are all real English words and all wrong. That diagnosis, and the fix, is covered in voice to text accuracy with an accent.
Composing aloud is a different skill from reading aloud
Worth separating, because people who try dictation and hate it are often failing at the composition part while blaming the recognition part.
Reading aloud is easy. The words already exist. Composing aloud means holding a sentence in your head while your mouth delivers the previous one, and if you have spent twenty years composing through your fingers, that pipeline is not built yet.
Three things make it faster to build:
- Start with things you already know how to say. Replies to messages, meeting notes, explanations you have given before. Not the hardest paragraph in a document you are stuck on.
- Talk to a person, not a screen. Imagine explaining it to a specific colleague. Speech gets structurally better the moment it has an audience.
- Overshoot, then cut. Say more than you need and delete afterwards. Trying to compose tight prose in one spoken pass is what makes people freeze.
Expect a week before it feels natural, and expect the first three days to feel worse than typing. That curve is normal and it is short.
The environment does more than your technique
You can speak perfectly into a bad situation and still get bad results.
The category that matters most is not volume, it is other people's speech. Steady background noise like a fan or traffic is comparatively easy to handle. Another conversation nearby is genuinely difficult, because it is the same kind of signal you are producing. A television with dialogue on it is the worst home-office offender.
Room acoustics matter next. A hard-surfaced room with no soft furnishing sends a delayed copy of your voice back into the mic. Sitting closer helps here too, because it improves the ratio of direct sound to reflected sound.
Full details, including per-environment advice for cafes, open offices, and cars, are in fixing dictation in background noise.
The one thing speaking clearly will never fix
Here is the honest limit, and it is important because people spend weeks working on technique to solve a problem technique cannot touch.
Proper nouns and specialist vocabulary fail for a structural reason, not an acoustic one. Your colleague's name, your product's name, the drug, the statute, the framework, the town you live in. These are rare words competing against common words that sound similar, and the common word usually wins regardless of how beautifully you said it.
You can identify this error class in one look. If your errors are scattered homophones and dropped word endings, technique will fix them. If the same three names come out wrong the same way every single time, no amount of enunciation will help. That distinction is the core of why Mac dictation gets the wrong words.
Built-in dictation does not learn from your corrections. You can fix the same name a hundred times and the hundred and first will be identical. That is a design limit rather than a bug.
Voice Keyboard Pro's Smart Vocabulary is the structural answer: a personal dictionary with replacement rules where you add the terms you actually use, along with what they get mistaken for. The correction happens at the source, so the wrong word stops appearing rather than getting cleaned up afterwards. For most people, twenty entries covers the overwhelming majority of their recurring errors.
A five-minute drill
If you want to build the habits deliberately rather than absorbing them over months, this works.
- Baseline. Open a blank document and dictate one paragraph the way you normally would. Count the corrections needed.
- Phrases. Dictate the same paragraph again, this time in complete clauses with pauses only between them. Count again.
- Distance. Same paragraph, same phrasing, but with the mic noticeably closer. Count again.
- Punctuation. Same paragraph with every mark spoken. Note how much cleanup disappears.
- Classify what is left. Whatever errors survive all four passes are almost certainly vocabulary. Write them down; that list is your dictionary.
Counting corrections on the same text is the only measurement method worth using here. Impressions are unreliable, because a single infuriating error colours a whole session that was otherwise fine.
Common mistakes, collected
- Speaking word by word. The biggest one, and the one that feels most helpful.
- Getting louder instead of getting closer. Distortion is not clarity.
- Trailing off at the end of sentences. The last words are the ones you will be fixing.
- Repeating a misheard word louder and slower. This almost never works. Say the whole phrase again normally, or fix it as text.
- Dictating with the phone at your waist. Distance and off-axis position at the same time.
- Bluetooth headsets in call mode. Many wireless headsets drop to a much lower-quality audio path when the mic activates. If your headset sounds thin when you dictate, that is why.
- Blaming the engine for a settings problem. A language variant mismatch looks exactly like poor recognition and is fixed in thirty seconds.
Frequently asked questions
Should I speak slower for dictation?
No. Speak at your normal conversational pace. Slowing down separates words from the context that helps identify them. If you are getting errors, change your phrasing and your mic distance before you change your speed.
Does an accent affect dictation accuracy?
Less than most people fear, and the fix is a setting rather than a performance. Match your language variant to how you speak. Do not attempt a different accent, which introduces inconsistency and makes results worse.
Is it better to whisper or speak normally in an open office?
Normally, but quietly, with the mic much closer. Whispering removes the vocal cord vibration that carries a lot of the information in speech, so whispered dictation is unreliable regardless of how close the mic is.
Why does dictation get worse when I concentrate harder?
Because concentrating usually means over-articulating and slowing down, and both hurt. The counterintuitive answer is to relax and talk the way you would to a colleague standing next to you.
Do these rules apply to iPhone as well as Mac?
All seven apply on both. The only meaningful difference is that phones tend to be held at variable distances, so consistency of position matters more. On a phone, Voice Keyboard Pro's Voice Edit also lets you speak a correction instead of placing a cursor with your thumb, which is where most phone-dictation frustration comes from.
What good technique is actually worth
An adult types at roughly 40 words per minute, a strong professional typist reaches 80 to 100, and everyone speaks at 130 to 150. The speed advantage of dictation is not in dispute. What determines whether you keep it is how much time you spend fixing the output.
Good technique is the difference between a paragraph that needs one small edit and a paragraph that needs six. That gap is what makes people quit, and it is closed by phrasing, distance, and a dictionary. None of it requires speaking like a robot.
Voice Keyboard Pro has a free tier with daily limits, which is enough to run the five-minute drill and see where your own errors actually come from. Pro is $4.99 a month or $34.99 a year, covering the Mac menu bar app and the iPhone keyboard.