← All posts

Is dictation faster than typing? The numbers

Yes, for most people, and by a wide margin. Typists in the largest study we know of averaged about 52 words per minute. People in conversation speak at roughly 160 to 200. Speech recognition on a phone has measured three times faster than the phone keyboard. For a long prompt to Claude Code, Cursor or Codex, that gap is minutes a day.

The honest answer has a second half, though. Raw speed is not where the time goes on a prompt. This post walks through the numbers, then where speaking wins, where typing still wins, and how to measure it for yourself.

How fast people type

The best data on typing comes from a 2018 study by Dhakal and colleagues at Aalto University, presented at ACM CHI. They recorded 136 million keystrokes from 168,000 volunteers who took an online typing test.

The numbers:

  • Average speed: 51.56 words per minute (standard deviation 20.2).
  • The slowest tenth typed under about 26 WPM. The fastest tenth typed over about 78 WPM.
  • The very fastest reached 120 WPM or more.
  • Average uncorrected error rate: about 1.2 percent.
  • 72 percent of participants had taken a typing course.

Two things to keep in mind. Participants were copying sentences shown on screen, which is faster than composing your own words. And a self selected group that takes a typing test online probably types more than average. So 52 WPM is a generous figure for writing a prompt from scratch.

How fast people talk

Mark Liberman at the Linguistic Data Consortium measured speaking rate in Switchboard, a corpus of 2,438 recorded English telephone conversations.

  • Counting all words over total conversation time: about 196 WPM.
  • Counting each speaker over their own turns: about 164 WPM.
  • With pauses removed, the net rate was about 236 WPM.

Dictating a task is probably slower than chatting, because you stop to think. Even so, the conservative 164 WPM figure is three times the average typist.

What happens when you add speech recognition

Speed of talking only matters if the machine keeps up. In 2016, researchers at Stanford, the University of Washington and Baidu tested this directly. In the study, 32 people entered everyday phrases on an iPhone, half in English and half in Mandarin, either with the built in keyboard or with Baidu’s Deep Speech 2 recognizer.

In English, speech was 3.0 times faster than the keyboard, with a 20.4 percent lower error rate. In Mandarin it was 2.8 times faster, with a 63.4 percent lower error rate.

The catch for a developer: that was a phone keyboard, and short phrases. A Mac keyboard in the hands of someone who types all day narrows the gap. It does not close it.

The arithmetic for one prompt

Take a 150 word prompt: the goal, the files involved, two constraints and how to check the result. That is a normal size for a task you hand to an agent.

150 words / 52 WPM (average typist)   = about 2 min 55 s
150 words / 78 WPM (fast typist)      = about 1 min 55 s
150 words / 164 WPM (speaking)        = about 55 s

Add 15 to 20 seconds to read the dictated text before you press Enter. Speech still beats a fast typist by about 40 seconds, and an average typist by close to two minutes. Over twenty prompts a day, that is somewhere between 13 and 35 minutes.

Where the time really goes

The bigger effect is not the minute you save. It is what you put in the prompt.

When you type, you edit as you go, and you cut context because it costs keystrokes. “Fix the upload bug” is a typed prompt. When you talk, you say the background you would give a colleague: which file, what you already tried, what must not change. Coding agents do better with that context, so the spoken prompt often saves a second round trip, which costs far more than the typing did. We go through how to speak a task clearly in how to dictate prompts into Claude Code.

Speaking has two real costs.

Cleanup. Spoken text is full of “so”, “um” and “no wait, I mean”. An agent reads all of it. Either you edit it out, or a tool does it for you.

Identifiers. A speech model that has never seen useSessionStore or stripe-webhook will spell them as ordinary words. Fixing those by hand eats the time you saved.

Typing still wins for short, exact input: a one word reply, a shell command, a regex, a single identifier. Use your hands for those and your voice for anything that takes a paragraph to explain.

Measure your own numbers

Your speed matters more than an average. A two minute test:

  1. Pick a task you were about to hand your agent.
  2. Time yourself typing the prompt in any editor. Copy it.
  3. Count the words:
pbpaste | wc -w
  1. Do the same task by dictation, read it once, fix what needs fixing, and stop the clock when it is ready to send.

Compare words per minute, and compare the prompts. Most people find the spoken one is both faster and longer, in the useful way.

Where Akova fits

Akova is a Mac dictation app built for that second test. You hold a hotkey, talk through the task, and Prompt mode turns the ramble into a structured prompt at your cursor in any terminal or editor. Custom vocabulary covers the identifier problem: add your repo’s names once and they come out spelled right. It runs on your Mac.

Clean up, custom vocabulary and the raw transcript are free. Prompt mode is part of Lifetime, and every Pro feature is free for 7 days first. The details are on the pricing section.

Run the timing test above on your next prompt. If speaking is not clearly faster for you, keep typing. If it is, you now know by how much.

Sources

Checked on 18 October 2026.