How to dictate an email instead of typing it

To dictate an email well, say the whole thing in one pass with the point near the top, and correct your slips out loud rather than stopping to fix them. Leave the tidying and the tone for afterwards, once the words exist.

Speaking is quick. The slow part is everything people try to do while they speak.

The point goes right after the greeting

Say why you are writing in the sentence straight after the greeting. Spoken drafts drift toward their point, and an email that finds its point in the third paragraph makes the reader do the finding.

That sentence says what they should do once they have read it. Maybe it is a yes by Friday. Maybe it is nothing, and they just need to know the files are late. Whatever it is, it is also most of a good subject line.

Everything after it is support: the date, and the one detail they would ask about if you left it out. If you cannot say that first sentence yet, think out loud first, and dictate after.

Short sentences, with a pause between each

Talk in short sentences and leave a beat between them. A pause is the clearest signal you can give anything transcribing you that a sentence has ended.

Speech runs on "and", and the transcript comes back as a single sentence that fills the screen. If a sentence has two "and"s in it, it wanted to be two sentences.

Short sentences are cheaper to fix, too. A wrong word in a long one sends you back to the start to work out what it was attached to.

Say corrections in the same shape as the mistake

Correct a slip by saying a marker phrase and then the right version, built the same way as the wrong one: "by the 10th, no wait, by the 8th". Whoever tidies the transcript later, you or software, can then see exactly which words go.

• Use a phrase nobody says by accident, like "no wait" or "scratch that".
• Repeat the words you are replacing, so "by the 8th" rather than a bare "no, eight" that leaves the reader guessing what the eight replaces.
• If a whole sentence went wrong, say "scratch that", pause, and say it again from the top.
• Correct once and move on, because a third attempt at one sentence usually means you had not decided what it says.

Keep your eyes off the screen while you talk

Look away from the screen while you dictate, and read nothing until you have finished. Watching the words arrive turns you into an editor halfway through a sentence.

The words land a beat late and sometimes wrong, so your eyes catch a bad one and your mouth stops to deal with it. Of all the dictation habits, watching is the one I would break first.

Look at the window instead, or close your eyes. The tidying gets its own pass afterwards.

Check names and numbers by eye

Read every name and number in a dictated email letter by letter before you send it. Transcription mistakes in those spots tend to sound right, so reading the draft back in your head lets them through.

I learned how loosely speech gets spelled from the two words that end dictation in the extension I make. The phrase is "End Flow", and the speech engine has written it down as "and flow", "Enflow", "anflow" and "EnFlo". So the code matches the shape of the phrase across all of those spellings, with a guard so that someone saying "ebb and flow" does not stop their own dictation. A surname you say once gets none of that help.

• Say dates with the day attached, "Thursday the 8th", so a mismatch is visible to you and to them.
• Type an unusual surname in after you finish, instead of spelling it aloud mid-sentence.

Choose the tone once the words exist

Say the email plainly, the way you would say it to them on a phone call, and decide how formal it should be afterwards. Trying to sound professional while you talk is how people end up dictating "I am writing to inform you that" to someone who sits at the next desk.

Tone is easier to judge when you can see the words. Out loud, you are deciding how firm to be before you know what you are being firm about.

With "Start with Flow" switched on in desktop Chrome, Unvent splits the job the same way. Ctrl+Shift+U in an email starts listening. You talk, then say "End Flow". The ums and repeated words go, a take-back keeps only the corrected version, and capitals and punctuation get fixed. Then it writes three drafts, Polished, Direct and Warm, and one key applies the one you want.

Chrome turns your speech into text on your own computer, and no audio is recorded or sent. The text then goes to Unvent to be tidied up and turned into the drafts, and none of it is stored or used for training.

Here is a made-up example, spelled the way the speech engine writes speech down, one line per pause:

hi Dana, so um i went through the proposal last night
and honestly it's it's really strong
my only worry is the timeline, we'd need the design files by the 10th no wait the 8th
uh could we talk it through on Thursday

What lands in the email:

Hi Dana, so I went through the proposal last night and honestly it's really strong. My only worry is the timeline, we'd need the design files by the 8th. Could we talk it through on Thursday?

It is still your wording: "so" and "honestly" survive. Changing the words is the drafts' job.

Direct reads: "Hi Dana, the proposal is strong, but we need the design files by the 8th to meet the timeline. Can we talk on Thursday?" Warm reads: "Hi Dana, I went through the proposal last night and it is really strong. I am just concerned about the timeline since we need the design files by the 8th, so could we talk it through on Thursday?"

Direct is for a colleague who wants the facts. Warm is for someone who spent a week on that proposal.

Type the emails that have to be exact

Dictate the emails whose main job is to get sent, and type the ones whose main job is to be exact. The overdue reply and the thank-you note you keep meaning to write belong in the first group, where any honest version beats none.

An offer belongs in the second, and so does a promise with a date in it. Those get read twice, and the second time by someone looking for a mistake.