The speed claim behind every dictation product traces back to a small set of real numbers. They are worth separating from the marketing, because the honest version is still remarkable, the dishonest version poisons trust in the whole category, and there is a well-known study that appears to contradict the whole idea. That study deserves a real answer rather than being left out of the article, so it is in here.
The numbers, with their sources
| Measure | Figure | Source |
|---|---|---|
| Conversational speech | 120 to 160 WPM | Established speaking-rate range |
| Speech input vs keyboard | About 3× faster | Ruan et al., Stanford, 2016 |
| Transcription typing | About 40 to 60 WPM | Standard typing-test range |
| Composition typing | About 31 to 40 WPM | The rate that applies to writing |
| Knowledge-worker time in email | 28% of the workweek | McKinsey Global Institute, 2012 |
| Time in work email daily | About 3.1 hours | Adobe Email Usage Study, 2019 |
Speech: 120 to 160 words per minute. Ordinary conversational pace for English speakers, before any tool enters the picture. See words per minute for how the measure works and why a “word” is defined as five characters.
Speech input: about 3× keyboard speed, measured. The controlled comparison is Ruan et al. at Stanford, 2016, which tested speech recognition against keyboard text entry for English and Mandarin on mobile devices. Speech came out roughly three times faster for English, and produced fewer errors after correction. It is one study on touchscreen phones rather than a universal law about desktops, and it remains the anchor citation a decade later because it was cleanly designed.
Composing at a keyboard: roughly 31 to 40 WPM. This is the number that surprises people, and it is the hinge of the entire argument. Typing-test scores of 40 to 60 or more measure transcription: copying text that already exists in front of you. Writing is composition, producing words while simultaneously deciding what they should be. The deciding is not free, and it happens in pauses that the typing test never sees. Stated plainly: your typing speed is not your writing speed, and any comparison that uses your typing-test score is comparing the wrong thing.
The stakes: hours, not minutes. McKinsey’s estimate puts 28 percent of the knowledge workweek into reading and answering email. Adobe’s 2019 survey measured about 3.1 hours a day in work email alone. Writing speed is not a niche optimization; it is a meaningful slice of the working year.
The study that says the opposite
In 1999, a team at IBM (Karat, Halverson, Horn, and Karat) published a study at CHI comparing speech recognition against keyboard and handwriting for both transcription and composition. Once error correction was included in the measurement, speech input came out slower than the keyboard, not faster. Raw entry by voice was quick. Fixing what the recognizer got wrong ate the entire advantage and then some.
That result was correct, it was influential, and it is the reason a generation of people tried dictation once and concluded it was a gimmick. It is also the most useful finding in this article, because it identifies the real variable. The bottleneck in dictation has never been how fast humans talk. It is the correction tax.
What changed between then and now is not human speaking rates. Two other things changed: recognition error rates fell by roughly an order of magnitude as neural models replaced the older statistical ones, and tools began doing the cleanup automatically instead of handing you a raw transcript and a correction interface.
So the honest framing is not “dictation is faster than typing.” It is: dictation is faster than typing to the exact extent that your tool removes the correction tax. That is a claim you can test, and it is what separates the tools in this category from each other far more than raw accuracy does.
What the raw numbers leave out
Dictating 150 words a minute of speech does not produce 150 words a minute of finished writing.
Raw speech is full of disfluencies: fillers, repeated words, false starts, and mid-sentence corrections. Every speaker produces them, including people who believe they do not. A raw transcript also arrives with approximate punctuation, no paragraph structure, and none of the formatting that written English needs to be readable. Numbers, times, prices, and units all come out spelled the way you said them rather than the way they are written.
If you spend three minutes cleaning up every five minutes of dictation, your effective rate collapses toward typing speed. That is the 1999 finding, reproduced in your own week.
This is why the only measure worth quoting is delivered words: finished, sendable text per minute of your attention, cleanup included. Everything else is a benchmark of a component rather than a measure of the job.
As one data point with its methodology stated: our own measurement across 500 or more real dictations, by one non-native English speaker, pauses and thinking included, comes to 124 delivered words per minute with the cleanup done by the tool. Against 31 WPM composition typing, that is the 4× figure on our homepage. It is our number, not a study, and it is the conservative end of the range rather than the flattering one: published speaking rates run well above 124, so a fast native speaker should expect to beat it.
Measure it on yourself in twenty minutes
Published averages describe a population you are not. The comparison that matters takes twenty minutes, and it is the only evidence that will actually change how you work.
- Pick a real task. Three emails you genuinely owe, or a page of notes you actually need. Not a typing test, not a passage to copy. Composition is the thing being measured.
- Type the first half. Start the timer when you start thinking, not when you start typing. Stop when the text is something you would actually send.
- Dictate the second half. Same rule for the timer. Include every second of fixing what came out wrong.
- Count the words in each, and divide by the minutes. Both numbers are now delivered WPM, which is the comparable unit.
- Repeat on a different day. The first dictation session of your life is not representative of anything; there is a real learning curve in the first week, mostly around learning to speak in finished sentences.
Two things usually show up. The dictated half is faster than people expect and longer than they expect. Dictated prose runs wordier, because speech is wordier. Whether that matters depends entirely on what you are writing.
Where typing honestly wins
Dictation is a first-draft and reply engine. It is not a universal replacement for the keyboard, and treating it as one is how people end up abandoning it.
- Editing. Reworking existing sentences by voice is slower than cursor and keys, even with voice-edit commands. The keyboard is a precision instrument and editing is precision work.
- Precision surfaces. Spreadsheets, design tools, code refactors. Keyboard territory, and not close.
- Shared spaces. Open offices, cafés, shared homes, and trains. This is the most common practical limit on dictation and no amount of software fixes it. If you cannot speak freely, none of the arithmetic in this article applies to you.
- Dense argument. Text where every sentence has to carry weight often benefits from the slower pace of typing. Many people who dictate seriously use both: dictate the draft, type the edit.
- Thinking styles. Some people compose better through their fingers, genuinely and not as a habit to be corrected. If that is you at your best, the arithmetic below does not apply, and that is a legitimate answer.
The arithmetic of switching
The savings scale linearly with how much you write, which makes the whole thing easy to estimate for yourself.
| Words per day | At 31 WPM typing | At 124 WPM dictation | Saved per day | Saved per year |
|---|---|---|---|---|
| 500 | 16 min | 4 min | ~12 min | ~1 working week |
| 1,000 | 32 min | 8 min | ~24 min | ~2.5 working weeks |
| 2,000 | 65 min | 16 min | ~48 min | ~5 working weeks |
| 4,000 | 129 min | 32 min | ~97 min | ~10 working weeks |
Yearly figures assume roughly 250 working days and a 40 hour week, and they are rounded, because pretending to precision here would be exactly the dishonesty this article is arguing against. Two thousand words a day is a normal volume for someone whose job includes email, messages, and notes. The calculator on our homepage runs the same arithmetic on your own numbers with its assumptions on display.
Treat these as an upper bound on the theoretical gain, not a promise. The real return is whatever survives your correction tax, your working environment, and the share of your writing that is actually composition rather than editing.
The fair summary
The speed gap between speech and composition typing is real, large, and stable across studies. The 1999 result does not contradict it; it measures the same gap and then subtracts a correction cost that has since fallen dramatically. Whether you keep the gap in your own work depends almost entirely on how much cleanup your tool does for you, which is the problem the current generation of dictation software exists to solve, with meaningfully different answers from each product.
For how they compare on that specific job, see the 15 best dictation apps for Mac. If where your audio is processed matters as much as how fast it comes back, the best private dictation apps covers verifying that claim yourself.
Sources: Ruan, Wobbrock, Liou, Ng, and Landay, “Speech Is 3x Faster than Typing for English and Mandarin Text Entry on Mobile Devices” (Stanford, 2016), later published in peer-reviewed form; Karat, Halverson, Horn, and Karat, “Patterns of Entry and Correction in Large Vocabulary Continuous Speech Recognition Systems” (IBM, CHI 1999); McKinsey Global Institute, “The Social Economy” (2012); Adobe Email Usage Study (2019). Our 124 WPM figure is our own measurement and is labeled as such wherever it appears.