TL;DR
- Speech ran 2.93 times faster than a phone keyboard in the one controlled study we could verify.
- Speech left 1.30% uncorrected errors in final text, against 0.79% for typing.
- Speech made fewer corrected errors during entry (5.30% vs 11.22%), so the errors you see are different.
- The study used short messages, touchscreen phones and an older engine. Test your own text before you decide.
153 vs 52
Words per minute, speech vs keyboard entry on touchscreen phones
Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones (arXiv:1608.07323)2.93 times
Speed of speech entry relative to keyboard entry in the study
Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones (arXiv:1608.07323)5.30% vs 11.22%
Corrected errors during entry, speech vs keyboard
Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones (arXiv:1608.07323)1.30% vs 0.79%
Uncorrected errors in final text, speech vs keyboard
Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones (arXiv:1608.07323)Dictation leaves more errors in your final text, while typing leaves more errors you fix as you go. The one controlled study we could verify found speech 2.93 times faster than a phone keyboard, with 1.30% uncorrected errors against 0.79% for typing.\n\nThose two numbers are the headline. The rest of this page covers what they mean, what they leave out, and how to measure your own dictation accuracy on a Mac.\n\n## What the controlled study actually tested\n\nMost hard numbers on this topic come from one paper. It is titled Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones. Ruan and colleagues wrote it. The arXiv record lists it as submitted in August 2016 and revised in January 2018. It was published in Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies in December 2017.\n\nThe title already tells you what the test covered.\n\n- Short messages. Participants wrote short texts, not reports, contracts or long emails.\n- Touchscreen phones. The keyboard baseline was a phone keyboard, not a Mac keyboard with shortcuts.\n- An older speech engine. The engine predates today's neural recognition, so its mistakes may not match what a current tool produces.\n\nWe could not verify the participant count. The text we could read did not state it, so this page leaves it out. Keep that in mind before you quote the study in a meeting or a memo.\n\n## Speech vs typing error rates side by side\n\nHere are the three numbers that matter most, in one table.\n\n| Measure | Speech entry | Keyboard entry |\n|---|---|---|\n| Entry speed | 153 words per minute | 52 words per minute |\n| Corrected errors during entry | 5.30% | 11.22% |\n| Uncorrected errors in final text | 1.30% | 0.79% |\n\nSource: Ruan et al., arXiv:1608.07323.\n\nRead the table in two passes. The first pass is speed. Speech ran at 153 words per minute, against 52 for the keyboard. The study puts speech at 2.93 times faster. If you want the broader speed question for everyday writing, our piece on whether dictation is faster than typing covers it.\n\nThe second pass is errors, and it is where most people get confused. Corrected errors are mistakes the writer spotted and fixed while entering text. Speech had fewer of them, 5.30% against 11.22%. Uncorrected errors are mistakes still sitting in the final text. Speech had more of them, 1.30% against 0.79%.\n\nSo speech did not win on accuracy in this test. It won on speed. The mistakes that matter to a reader are the ones that reach the final text, and speech left more of those behind.\n\n## Why dictation errors slip past you\n\nA typing error is usually one wrong letter. A dictation error is often a whole wrong word that still makes sense in the sentence. That is why it is easier to miss.\n\nThree patterns show up often in everyday use.\n\nSound-alike words. You say "their" and the engine writes "there." Both are real words, so your eye moves on.\n\nNames and technical terms. A surname, a product name or a library name comes out as a common word that sounds close. Check each one every time.\n\nPunctuation and breaks. Speech often arrives without the commas and paragraph breaks you would type without thinking. A sentence runs on, or a list collapses into one line.\n\nWhy do you miss them? Once you speak a sentence, it feels finished. You read less closely than you would if you had typed each letter yourself. This is a habit, and you can change it.\n\n## Why typing errors get caught sooner\n\nTyping errors show up as you go. You see the wrong letter the moment it appears, and you press delete. That is why the corrected error rate for typing in the study, 11.22%, is roughly double the speech rate of 5.30%. Typists fixed more of their own mistakes during entry.\n\nTyping has its own traps. Autocorrect changes words you meant to keep. A tool can quietly fix a transposed pair of letters, and you never see the original. A sentence can also look clean on screen while one word is wrong.\n\nThe uncorrected typing rate was 0.79%. So typing still leaves errors behind, just fewer than speech did in this test.\n\n## What the lab numbers do not tell you\n\nThe study answers a narrow question. It does not tell you how dictation performs on your Mac, in your app, with your voice.\n\nFour limits matter most.\n\n- Length. Short messages leave little room for errors to pile up. A long report has more chances to go wrong, and this study did not measure that.\n- Content. Everyday messages use plain words. Unusual names and technical terms are harder for any engine.\n- Speaker. Accents, background noise and speaking pace change results. A study sample is not your voice or your room.\n- Engine. Current dictation tools use newer recognition than the study engine. The gap could be large or small. We do not know, and this page cannot tell you.\n\n## Dictation for names, code and numbers\n\nNames and numbers deserve their own look, because they do the most damage in a finished email.\n\nCode is harder still. Some tools let you say "open parenthesis," but you rarely want to. A typed symbol is faster and less likely to be misread. Use dictation for the prose around code, such as a commit summary or a comment, and type the code itself. If you work in Cursor, the voice assistant for Cursor on Mac page covers that workflow in more detail.\n\nNumbers are the other trap. A dictated date, amount or reference can come out as words, digits or a near miss. Read every number back, out loud, before you send anything. One wrong digit in a payment reference does more damage than a missing comma.\n\n## Mistakes people make when comparing the two\n\nMost comparisons go wrong before anyone starts. These are the errors people make most often when they judge dictation against typing.\n\nThe first mistake is measuring the wrong thing. People count how many words they spoke per minute and call it accuracy. Speed and accuracy are separate measures. The study reports both, and you should track both.\n\nThe second mistake is proofreading unevenly. You may read typed text twice and dictated text once. Give both versions the same review, or the comparison favors whichever one you checked harder.\n\nThe third mistake is judging from memory. Ask yourself which version felt cleaner and you will guess. Write the error counts down while you work, so the comparison rests on paper, not on feeling.\n\nThe fourth mistake is ignoring where errors get found. Typing errors usually get fixed in the moment. Dictation errors can surface later, when a colleague points one out. Note where each error was found, not just how many there were.\n\n## A quick test you can run this week\n\nRun a small test on your own work before you change your habits. Lab numbers are a starting point, not a verdict.\n\n1. Pick a real message you need to write. Keep it to one paragraph, such as a reply to a client.\n2. Type the paragraph in your usual app, such as Mail or Pages. Proofread once the way you normally would. Write down each error you still find.\n3. Dictate the same paragraph into the same app. Use the same content and tone. Do not copy the typed version.\n4. Read the dictated version aloud, slowly. Fix every error you hear or see, then write down the count.\n5. Compare the two counts. Only errors that survive your normal proofread count.\n\nThree mistakes ruin this test. Testing with one sentence gives you noise, not a pattern. Reading the dictated text quickly makes you miss sound-alike words. And changing the content between the two versions makes the comparison meaningless.\n\nFor a rough baseline before the full test, try the dictation accuracy checker and the typing speed test on this site. Then run the paragraph test on the message you actually need to send.\n\n## Where each dictation tool stands on Mac and Windows\n\nYour error rate depends on the tool, its setup and its requirements. Here is what the vendor pages say.\n\n| Tool | Runs on | What the vendor page says | How to start |\n|---|---|---|---|\n| Apple Dictation | macOS | Turn it on under System Settings > Keyboard > Dictation, then click Enable. Voice Control replaces it when on. | Set in System Settings |\n| Microsoft Word Dictate | Word for Microsoft 365 on Windows and Mac | Needs a Microsoft 365 subscription, a microphone and a reliable internet connection | Home > Dictate, or Option+F1 on Mac |\n| Windows voice typing | Windows 11 | Needs internet access and uses Azure Speech. Supports 46 language variants on Windows 11. | Win+H |\n\nSources: Apple Support and Microsoft Support, linked in the source list below.\n\nTwo details stand out. Word's Dictate needs a subscription and a reliable connection before it works at all. Windows voice typing also needs internet access, and its fluid dictation feature, which corrects grammar and punctuation, is a Copilot+ PC feature available only in English locales.\n\nOn macOS 27, Apple says Apple Intelligence dictation uses an on-device model to improve spelling, punctuation and capitalization. Apple lists this for English on Mac models with M3 or later and at least 12GB of unified memory. Dictation is not available in every language or region. Check your version under Apple menu > About This Mac before you follow any menu path here, because older macOS versions may look different.\n\nFor a side-by-side look at the built-in option and the AI dictation apps, read Apple dictation vs AI dictation apps.\n\n## Ways to cut dictation errors on your Mac\n\nHere is what helps in practice.\n\n- Dictate in short chunks. Speak one or two sentences, then stop and read. Long runs hide errors.\n- Check names the first time. Write a test note with your names and product terms. See how the engine spells them, then fix the rest by hand.\n- Say punctuation out loud if your tool supports it. If it does not, add commas and periods after you finish. Check where the engine placed them.\n- Read aloud before you send. Your ear catches sound-alike words that your eye skips.\n- Use a personal dictionary if your app has one. Add the words you use every day, such as client names or internal terms.\n- Set the language first. Open System Settings > Keyboard > Dictation and confirm the language before you start. A wrong language setting causes errors fast.\n- Control the room. Keep the microphone at a steady distance and the background quiet. Noise is an error source you can control.\n\n## Where Lazur fits\n\nIf you dictate every day, Lazur is one option to test. Hold fn to dictate clean, formatted text at the cursor in any app. Lazur also keeps snippets, a personal dictionary and clipboard slots. Those help with the names and phrases from the section above.\n\nDictation is included in both paid plans and is unlimited on each. Pro costs $18 a month, or $15 a month billed annually. Power costs $120 a month, or $96 a month billed annually. The 3-day Pro trial starts when you first open the app, and it needs no credit card.\n\nWe do not have a verified accuracy number for Lazur. Run the test above with it before you decide. You can start on the download page, and the pricing page lists both plans.
FAQ
Is dictation more accurate than typing?
Not in the controlled study we could verify. Speech made fewer corrected errors during entry, but left more uncorrected errors in final text: 1.30% against 0.79% for typing. Results depend on the engine, the speaker and the text.
Which is more accurate for emails, speech or typing on a Mac?
We found no verified study on email specifically. Run the five-step test below on a real email. Count the errors that survive your normal proofread, then compare the two counts.
Does Microsoft Word dictation need an internet connection?
Yes, according to Microsoft's Word dictation page. It needs a microphone, a reliable internet connection and a Microsoft 365 subscription. Start it from Home > Dictate, or press Option+F1 on a Mac.
Can I trust the 2016 numbers for dictation today?
Treat them as a baseline, not a forecast. The study tested short messages on touchscreen phones with an engine that predates current neural recognition. A test with your own text tells you more.
Sources
- Comparing Speech and Keyboard Text Entry for Short Messages in Two Languages on Touchscreen Phones (arXiv:1608.07323) · arXiv / ACM IMWUT · checked 2026-11-06
- Dictate your documents in Word (Microsoft Support) · Microsoft · checked 2026-11-06
- Use voice typing to talk instead of typing on your PC (Microsoft Support) · Microsoft · checked 2026-11-06
- Use Dictation on Mac (Apple Support) · Apple · checked 2026-11-06


