The time dictation actually saves
Speaking beats typing, but the bigger win is the editing you never do: the pass that disappears when text arrives finished at the cursor. With worked sums.
Dictation saves time in two places, and the second one is worth more. Speaking is faster than typing; that part is obvious. The part people underestimate is the editing pass that never happens, because the text was already finished when it landed in the field. Plenty of dictation tools miss this. They swap typing for proofreading and call the difference a saving.
The one number that holds up
There is surprisingly little solid research on speech versus keyboard. The best-known study comes from Stanford, the University of Washington and Baidu in 2016. In English, speech input was roughly three times faster than typing, with an error rate 20.4 percent lower; Mandarin showed a comparable margin. (Paper, Stanford’s summary)
Two caveats belong with that number. It was measured on phones against a touchscreen keyboard, not against a decent desktop one. And the task was dictating short, prescribed sentences, not writing an email to a client where you change your mind halfway through the paragraph.
The rule of thumb everyone repeats, 150 spoken words a minute against 40 typed, is not a study. It is fine as an order of magnitude. It is not evidence, and we are not presenting it as any.
The real work sits between transcript and finished text
Say a paragraph out loud and look at what comes back. Spoken language: a filler at the start of the sentence, an abandoned clause, a correction mid-flight (“no, wait, I mean Thursday”), a list that was bullets in your head and one long sentence in the transcript. Typed out verbatim, none of that is ready to send.
Do that cleanup yourself and you hand back most of the head start. In Skald a cleanup model does it, and it knows where you are writing: which app is in front, what the window is called, what role the focused field plays. A Slack DM stays a DM. A formal email gets a greeting and a sign-off. In an editor your identifiers are left alone and only the spoken punctuation is resolved.
The boundary is strict. Names, numbers, dates and technical terms are never added. If you did not give a reason, Skald does not invent one. Proper nouns your microphone keeps mangling go into the dictionary once.
Three sums from an ordinary working day
The numbers below are assumptions, not measurements. Substitute your own.
| Task | Typed | Spoken | Difference |
|---|---|---|---|
| Chat reply, ~40 words | ~1 min | ~15 s | ~45 s |
| Client email, ~180 words | 6–8 min | ~2 min incl. a read-through | ~5 min |
| Ticket or doc paragraph, ~300 words | 10–12 min | ~3 min | ~8 min |
A day with eight chat replies, three emails and one longer piece: six minutes, plus fifteen, plus eight. Call it half an hour. Over a working week that is roughly two and a half hours, over a working year around a hundred.
The sum is only as good as its assumptions, and it is unfair in one direction. It ignores that spoken sentences often land on the first attempt, because you do not restart them twelve times the way you restart a typed one. It also ignores the days you record something twice.
Where dictation buys you nothing
In an open-plan office with nowhere to go. On “ok” and “sounds good”. On code with more brackets than words. And whenever you do not yet know what you want to say. Dictation speeds up phrasing, not thinking.
For that last case there is Brain-Dump: a window of its own that you talk into for ten minutes and that hands back structured Markdown. It will not think for you. It will sort what you thought.
If you write in Word, it has a built-in Dictate button too. How to turn it on, and where it falls short, is in the guide to dictating in Word.
Run the numbers yourself
What this comes to at your own writing volume is laid out on the time-savings page. Skald runs on macOS 13+ and on Windows 11 for both x64 and ARM64, get it here. The free plan gives you 50 dictations a month with no card, Pro is €9 a month with a seven-day trial, and every tier is listed on pricing.