Dictation when typing hurts

What dictation actually changes

Most of what a knowledge worker types is prose: email, messages, documents, notes, commit messages, tickets, code comments. Prose is the part dictation handles well, and for many people it is the majority of their daily keystrokes. Speaking it instead of typing it removes those keypresses entirely.

What it does not remove is everything else — navigating, selecting, clicking, keyboard shortcuts, and writing code itself, which is punctuation-dense and does not dictate comfortably. Anyone telling you that voice input eliminates keyboard use is describing a different category of tool, and that tool exists: see Talon below.

What the measurements show

The direct experiment is the one worth knowing about. When researchers put electromyography sensors on people doing text work, replacing a substantial share of keyboard and mouse input with speech recognition reduced static muscle activity in the forearm and neck, reduced it to some extent in the shoulder, and increased the pauses in forearm and shoulder activity.^[1] Those are the regions computer-related upper-limb complaints centre on.

The epidemiology points the same way about exposure. Across the literature on computer work, the duration of computer use is associated most consistently with hand and arm symptoms rather than neck and shoulder ones^[2] — and hand-and-arm exposure is precisely what dictation displaces. Keyboard intensity has been studied enough to fill a review of 558 publications.^[3]

The honest limit, stated because leaving it out would make the rest worth less: a systematic review of computer work and clinically verified disorders found the evidence limited.^[4] Reducing a known exposure is a reasonable thing to do about pain. It is not the same as a proven treatment, and we are not offering one.

You also write more, and better

In real clinical records, dictated notes averaged 590 words against 378 typed — about 56% longer — with a larger vocabulary.^[5] A later controlled study with experienced physicians found the same shape and measured more of it: 320.6 words dictated against 180.8 typed, 170.9 unique words against 120.4, fewer uncorrected errors (1.5 against 2.9 per note), and higher assessed quality (7.7 against 6.6).^[6]

The mechanism is not mysterious. Typing asks you to convert each word into a sequence of finger movements while also finding keys, tracking the cursor and repairing slips. Speaking does not, so more of your attention stays on what you are actually trying to say. The practical effect is that the note you would have written as four fragments gets written as a paragraph — which matters most on the days your hands hurt, because those are the days the writing gets skipped.

One keystroke in sixteen is a correction

A keystroke saved is a keystroke saved whether or not it was the right one, and a startling share of them are not. Across 136 million keystrokes, about 6.3% were corrections — one in sixteen.^[7] In the one controlled study that measured both input methods together, the keyboard’s corrected-error rate was 11.22% against speech’s 5.30%.^[8]

For this page that is not a quality argument, it is an arithmetic one: more than twice the mid-entry correction work means more than twice the backspaces, and every one of them is a keypress your hands made. We publish our own word error rate and the method behind it, which is more than the keyboard has ever been asked for.

What to look for, whichever tool you choose

• It types into the app you are already in. A tool that puts text in its own window and asks you to copy it out has handed the keystrokes back.
• A hotkey you can press without strain. Anything you press hundreds of times a day should be remappable to something comfortable — a foot pedal, a mouse button, a single key.
• It handles punctuation on its own. Saying “comma” and “full stop” out loud all day is exhausting and slow.
• Low latency. If the text lags several seconds behind your voice you will start typing again out of impatience, which defeats the purpose.
• It runs on your operating system. Obvious, and still the point where most options fall away — see our Linux page.