Voice-to-text as a thinking tool: talk before you type

Voice-to-text works best as a thinking tool, not a typing replacement: you talk through the messy version of what you mean at talking speed, then edit the transcript into the message you send. The talking is the thinking. The typing comes after.

Dictation has disappointed people for decades because it promises finished text and delivers a mess. Treat the mess as a first draft instead, and the same transcript becomes a small miracle: a draft that exists, seconds after you had nothing.

You already draft out loud

Most people compose their best drafts out loud, by accident. You have been circling a difficult email all morning, a coworker asks what is wrong, and your answer is the email. One sentence, said in one take, clearer than anything you typed.

Then you turn back to the keyboard and the sentence dies. You add a warm-up paragraph. You soften the point until it needs a second read to find, and the email gets longer and worse.

The difference is the listener. Speech comes with a person attached, so you instinctively lead with what matters and stop when they have it. A text box is not listening, and nothing pulls the point to the front.

Voice-to-text, used as a thinking tool, is a way of keeping the listener. You are telling someone, and the transcript is a side effect.

Speech has no backspace

Talking unsticks a stuck message because speech has no backspace. I have said before that nobody has writer's block in a chat box, and I stand by it: the words show up. What stalls the hard message is the delete key. You type a version, judge its tone, and erase it before it can grow into the right one. Twenty minutes of this and the box is empty. The words showed up; the delete key ate them.

Out loud, the clumsy version gets said. You hear that it is wrong, you say "what I mean is", and the better version follows. The bad sentence turns out to be the step the good sentence stood on. Typing lets you destroy the step before it can hold your weight.

The messages you dread are the best candidates for talking first. The same anxiety that makes the send button hard to press sits on your typing hands, and it has much less grip on your voice. One take, door closed. It will come out imperfect, and it will come out.

How to talk a draft worth editing

The way to get a transcript worth editing is to talk as if the reader were in the room. Before you start, say out loud who it is for and what you want from them. Then keep a few habits:

• Open with the ask, the way you would if the reader were standing there.
• When a sentence comes out wrong, do not stop. Say "what I mean is" and keep going.
• Stop when you run out of things to say. It will never sound finished, and that is fine.
• Stand up if you can. Pacing keeps the sentences coming. I cannot tell you why, only that it does.

The first try feels like leaving a bad voicemail. The second try feels like thinking. Typing became the default way to draft because it is quiet in an open office, and that is the whole reason. If talking to a computer looked less silly, half the first drafts written today would be spoken.

The transcript is a draft, not the message

Raw speech should never be sent as it is; it is material for an edit, and the edit is fast. Speech written down looks rough in a way it never sounds. When I was building the voice mode in Unvent I read pages of my own transcribed talk, and it was humbling: false starts, and the same word three times in one sentence. I sounded nothing like I thought I sounded.

The transcripts taught me something else, though. The point was always in there, and it was almost never at the beginning. I talked my way toward it, and somewhere in the last third it would arrive, plain and sendable. The sentence I would have agonized over at a keyboard simply fell out, wedged between two false starts. Spoken drafts have the goods. They just deliver them late.

Editing a bad draft is an easier job than writing a good one, and a different job. The thinking is done, so the edit is quick:

• Cut the repeats and the warm-up. Spoken drafts clear their throat out loud.
• Find the ask, which usually landed near the end, and move it to the top.
• Cut length. Speech runs long because talking is cheap.

Some of that cleanup is mechanical, which is why I taught Unvent's Flow Mode to do it as you speak: fillers, stutters, and self-corrections tidy themselves up before the words land in your draft, and the transcription happens on your device, so what you say stays on your machine. What is left for you is the human part: deciding whether the message says what you mean.

The draft you talk is the only first draft that arrives already sounding like you. Close the door. Say who the message is for, then talk until you run out. Read what you said. Somewhere near the end will be the sentence you spent all morning deleting, and this time it survived. Move it to the top and send it before the keyboard talks you out of it.