No medal for typing the email

The dentist emails. Can I move my appointment to the week after next? I can, and there are three things to say back: yes, Tuesday or Thursday, and please keep the morning slot. That is the whole message. It takes me eight minutes to type, because somewhere along the way I rewrite the opening line twice, wonder whether “thanks so much” is too much, and read the thing back as if it were going to be graded.

Nobody is grading it. Nobody has ever been grading it. There is no medal for typing the email, and I think most of us have been quietly behaving as though there is.

The transaction share

Take a look at your own inbox and sort the last fifty things you sent into two piles. The first pile is transactions: here’s the file, yes to the date, the two numbers you asked for, received with thanks. The second is the pile where the words themselves were the point: the letter, the apology, the pitch, the message to someone who matters where how you said it was the whole content.

My own split is somewhere around ninety to ten, and that’s a guess from my personal inbox, not a survey. But the data on how much of a working day goes on this stuff is not a guess. Microsoft’s analysis of its own users found the average person spends 57 per cent of their time communicating, in meetings, chat and email, and 43 per cent making anything. The heaviest email users put in nearly nine hours a week on email alone. A later Microsoft report counted the average employee receiving 117 emails and 153 chat messages a day. Most of those are transactions, and most of the replies are too.

Fig. 1 · The transaction share Most of what you write is a transactionTransactionalhere’s the fileyes, Tuesday worksthe two numbers you asked forreceived, thanksWorth hand-craftingthe letterthe apologythe pitchMy guess from my own inbox, not a survey.
The second pile is small, and it is the only one where the typing is the point.

Here is the claim. For the transactional pile, the value of your time is not in the typing. It’s not in the choice of “regards” over “best”, not in the comma, not in the fourth read-through. The value is in the meaning, and the meaning takes about fifteen seconds to say out loud. Everything after that is a tax we’ve agreed to pay because the keyboard was the only tool we had.

The lone author was always a committee

Part of why the tax feels compulsory is a story we tell about writing. The name on the book sat at a desk and chose every word. The name on the article did the research. The person who sent the email composed it.

It was never true. Milton was blind when he wrote Paradise Lost; he dictated it to whoever was in the room, his daughters among them. Dostoevsky, up against a contract deadline that would have cost him the rights to his own work, dictated The Gambler to a stenographer in twenty-six days, and then married her. Henry James, his wrist gone from decades of pen-work, dictated his last and best novels to a typist and found that his sentences got longer and stranger because he was speaking them. Churchill dictated nearly everything, walking up and down. Nobody reads those books and thinks less of them because a second person did the typing.

And it’s not just dictation. Books have editors, who change a great deal. Articles have research assistants and fact-checkers. Every piece of writing you’ve admired in the last fifteen years very probably went through a spell-checker, and probably a grammar tool too. The author’s meaning, the author’s name. Somebody else’s hands on a lot of the keys. We’ve always been fine with this for the people at the top of the trade, and oddly precious about it for ourselves at the bottom, with the dentist.

Talking is faster than you think

The other reason to stop typing the transactions is simple arithmetic. In 2016, researchers at Stanford, the University of Washington and Baidu sat forty-eight people down with an iPhone and had them enter the same phrases by keyboard and by voice. Speech was three times faster than typing in English, and nearly three times faster in Mandarin. That was speech recognition from a decade ago, and the lab timed people copying short phrases, which flatters both sides. A real message is composed, not copied, and that’s where my own numbers come in.

Fig. 2 · Five minutes talking The ratio from my own logSaid out loud5 minTyped25 to 30 min01530 minthe 2016 lab result: 3×My stopwatch, since summer 2026.
Five minutes of talking is twenty-five to thirty minutes of typing, on my stopwatch.

My own numbers, from a stopwatch I’ve run since the summer, are worse than the lab’s. A five-minute walk’s worth of talking comes out at twenty-five to thirty minutes of typing. Nobody has thirty minutes of typing in their fingers on a Tuesday evening; I have five minutes of talking in me pretty much always. Add it up and the time saved since I started measuring is now longer than the Iliad and the Odyssey read aloud back to back. My last book was dictated this way, most of it on the same loop past Holyrood, and driftcasting is the name I gave the habit.

I should say plainly that being able to do this is a privilege. It needs somewhere you can talk without annoying anyone, which means a room of your own, a walk, or a car. Not everyone has that, and some people can’t speak to a machine comfortably at all, or at all. If the only thing stopping you is feeling self-conscious, though, I’d push you to get over it, because the gap between people who talk to their computers and people who type at them isn’t a gap in speed. It’s a gap in what gets done at all. Thirty minutes of typing is a thing you put off. Five minutes of talking is a thing you do on the way to the shops.

Divergent you, convergent editor

So the method is: say the meaning, let the editor type it. But there’s a right way to do that and a way that produces something worse than the eight-minute email, and the difference is who’s doing the thinking.

When I dictate a reply, I’m not saying “write back to the dentist”. I’m saying what each line should carry: yes, two days that work, keep the morning. If it’s a longer piece, I’m saying what each paragraph is for and roughly in what order, and I’m doing it in my own words, badly, with the ums left in. That’s the divergent half: the ideas, the shape, the tone, all mine, all messy. The editor’s job is the convergent half: tighten it, format it, make it sound like me on a good day, and add nothing. Then I read it, which takes a fraction of the time typing it would have, and I cut anything I didn’t say.

Fig. 3 · Divergent you, convergent editor The meaning is yours; the typing isn’tnext oneYOU, WALKINGthe points,the order, the tone,what each paragraphcarriesTHE EDITORtightens, formats,matches your voice,adds nothingYOU, AT THE DESKread it,cut anythingyou didn’t say,sendadds a paragraph you didn’t sayworkslop
You do the divergent half, the editor converges, and anything it adds comes out.

That last rule is the one that matters, because the failure mode has a name now. Last year researchers at BetterUp and Stanford surveyed over a thousand desk workers and found that four in ten had received what they called workslop in the previous month: AI-written work that looks finished and says nothing, which the recipient then spends two hours decoding or redoing. People who sent it were rated less trustworthy, less capable, less reliable. The cost worked out at something like 186 dollars per person per month, which is a strange number to attach to politeness, but there it is.

Workslop is what you get when the editor is allowed to do the divergent half. The reply to the dentist comes back as four paragraphs with a warm opener and a bulleted summary. The two facts you had to send become a page. That isn’t saving anyone time; it’s moving your eight minutes onto someone else’s desk and tripling them. The test I use is blunt: if the recipient would rather have had the voice note, the editor has overstepped. A transactional reply should be shorter than what you said out loud, not longer. If the editor adds a paragraph you didn’t dictate, that paragraph goes, every time, however nice it reads.

When you should type it yourself

None of this touches the second pile. There are messages where the time is the message. The letter to someone you love. The apology that needs to be felt before it’s sent. The note to a friend who’s lost someone. Anything where the struggle to find the words is part of what you’re giving. You know which these are; they’re the ones you’d be slightly ashamed to have dictated, and that shame is a good instinct. Hand-craft them. Rewrite the opening line twice. That’s not the tax, that’s the point.

And even there, asking for help isn’t cheating. People have always said to a friend, “I want to tell her how I feel and I can’t get the words.” That’s what the editor is for in the hard cases too: not to write the letter, but to help you find the sentence you were circling. Just don’t send its first draft. We’re not all Wordsworth, and it’s fine. The meaning is still yours. The typing never was.

How I actually do it

  • Transactions get spoken, never typed. Yes, the date, the file, the two facts. Fifteen seconds of talking, a glance, send.
  • I dictate the paragraphs, not the request. What each one carries, in what order, in my words. “Write me an email about X” is how you get workslop.
  • Shorter than the dictation, or it’s wrong. For transactions, the reply should be tighter than what I said. If it’s longer, something got added.
  • Cut what I didn’t say. The review is the job now. Anything I don’t recognise as mine goes, however well it reads.
  • The second pile gets my hands. If I’d be embarrassed to have dictated it, I type it, and I take as long as it takes.

Further reading

All musings