Writing a Book by Dictation: Software and Workflow

You can write a book by dictation, and the software is the easy part. In a controlled study, English speech entered on an iPhone reached 153 words per minute, against 52 on the keyboard, which is 2.93 times faster ¹. Speed of speech is not what limits a book, though. Planning takes most of the work of composing, and recognition errors in a raw draft are frequent. The method that works is to outline scenes first, dictate one scene per session, clean the transcript, and then give the edit as much care as the drafting. This guide covers each step, for fiction and nonfiction alike, and separates measured results from what authors report about themselves.

Our overview of dictation for work and study covers the other professions, and the guide to dictation for academics and theses covers long academic documents. For the general method, see our guide to dictation.

Dictating is faster than typing, but the gain depends on practice with the method

Speech is faster than typing for getting words onto the page. The measured comparison is on a phone. Results from the same study give 52 words per minute on the phone keyboard and 153 by speech ¹, so an hour of continuous entry would produce about 3,120 words by keyboard and 9,180 by speech ^{2 3}. These are ceilings for sustained entry of short phrases in a quiet room. They leave out the thinking that a book needs.

The gap between speaking and composing is the important one. Composing by speech with earlier software was reported at 8 to 15 words per minute, which the authors describe as much lower than normal speaking speed ⁶. Speed in a book depends on how fast the next sentence is formed, so the speaking rate is only part of the answer.

Practice closes much of that gap. In a longitudinal study of speech software, users became more efficient at dictation, navigation and error correction as they gained experience ⁷. The largest productivity gains came from better dictation quality and from using navigation commands ⁸. Experience also matters in the IBM study of letter writing. People with years of dictation experience composed letters between 0 and 25 percent faster than novices, depending on how complex the letter was ⁹.

Author blogs and forum posts often report drafting quickly by dictation. We treated those reports as self-reports and did not use them as evidence. We found no controlled study that times book-length drafting by dictation, so the honest answer to "how many words per hour" is that the figure depends on how much of the hour is spent planning. For the full question, see how many words per hour you can dictate.

Writers have dictated books for centuries, from Milton to Henry James to Terry Pratchett

Dictation has a long history among writers. John Milton wrote Paradise Lost by dictation after he went blind in 1652, with the help of amanuenses and friends ¹⁰. The same workflow suits any writer who can speak more easily than they can type.

Henry James shows the same order of work in practice. His typist wrote that he worked out the scenes of a book in detail before he began dictating it ¹¹. The planning came first, and the speaking came after.

Terry Pratchett's late work shows the modern version. His assistant, Rob Wilkins, said in an interview that Pratchett left more than 20,000 words toward his autobiography, all of them dictated to him ¹². Wilkins, who was also his business manager, is describing his own work, so treat the figure as first-hand rather than independent. For the question of which authors did this, see famous authors who dictated their books.

An outline is what makes dictation work: you speak scenes you have already planned

Most of a book's time goes to planning, not to producing the sentences. In an IBM study of letter writing, about two-thirds of the time spent composing a letter went to planning, whether the letter was written, dictated or spoken ¹³. Dictation does not remove that share. It moves the sentence formation to the speaking stage, so the outline has to be ready when the session starts.

A workable method for a novel or a nonfiction book:

1. Write a scene list in plain text, one line per scene, with the point of the scene, the people in it and the ending.
2. Pick one scene per session. Speak it through in one pass, without stopping to fix sentences.
3. Save the transcript under the scene name, so the book can be rebuilt in order later.
4. Fix obvious errors in the same session, or leave them marked for the edit.
5. Leave the prose rough. The edit pass comes later, after the whole draft exists.

Apple says you can dictate text of any length on a Mac without a timeout ¹⁴, so session length is a matter of stamina. A pause is different: Mac Dictation stops by itself when it detects no speech for 30 seconds ¹⁵. Plan sessions by how your voice holds up, which is covered in the software section below.

The outline also makes the navigation commands useful. The gains in the longitudinal study came partly from using navigation commands ⁸, and a scene list gives you named places to move between.

Fiction needs a method for dialogue, character names and punctuation

Fiction has three hard parts for dictation: the speech tags and quotation marks in dialogue, the names of invented people and places, and the punctuation that carries rhythm. Each needs a deliberate method, and none of them is solved by the software alone.

Dialogue needs its own commands

On a Mac, the spoken phrase "quote" and "end quote" produce opening and closing quotation marks ¹⁶. "New paragraph" is the same as pressing Return twice, and "new line" is the same as pressing it once ¹⁷. The catch is timing. On a Mac, the new line or new paragraph you say during Dictation appears when you are done dictating, not as you speak ¹⁸. Check the layout of a dialogue passage after you stop, not while you speak it.

We did not find a study that measures recognition accuracy on dialogue as such. Treat the method as practice, and check every exchange in the edit.

Names are a weak point for recognition

Apple researchers report that end-to-end speech recognition still has a significant error rate for entity names that appear infrequently in its training data ¹⁹. Invented names, places and terms from your own world are likely to appear rarely in any training data, which is the pattern the study points to, so check them closely.

Some speech services let you add a phrase list to raise the weight of particular words. Azure AI Speech sets phrase list weight in a range from 0.0, which disables the list, to 2.0, the maximum influence ²⁰. Dragon stores custom words in each user profile ²¹. Apple's own Dictation pages, in the versions we read, did not describe a custom word list for general text, so check the current settings before you rely on one. Whatever tool you use, keep a names list at the top of the outline, and search for each name after the draft is complete. Our glossary entry on custom vocabulary explains the idea in more depth.

Punctuation is partly automatic

On a Mac, Dictation adds certain punctuation automatically as you dictate, in supported languages, and other marks need to be spoken ²². Test the setting on a page of dialogue before you start a long session, and read the first chapter aloud to see where the marks went. The terms for these behaviors are in our glossary entry on automatic punctuation and spoken punctuation.

Nonfiction dictates best as a talk: explain each section to one imagined reader

Nonfiction shares the planning problem with fiction, but the sentences are usually easier to speak. The method that works best is to explain each section aloud to one imagined reader, the way you would explain it to a colleague who has asked. You get the logic in order, and the reader gives the planning a clear target.

Numbers and symbols need care. Speaking a figure is easy, but the written form is not, so mark each figure in the draft and check it against its source during the edit. Academic and professional books have their own citation and format rules. Our guide to dictation for academics and theses covers those in more detail.

Walking helps generate ideas, but we found no study of walking while drafting

Does walking help you write? The one controlled study we found tested creative thinking, not writing. Adults who walked on a treadmill increased their scores on a divergent-thinking test, and walking raised the creativity scores of 81 percent of participants on the alternate uses test ²³. On a convergent test, the boost reached only 23 percent of participants ²⁴.

So walking may help you find the next idea for a chapter, and you can dictate those ideas into a note while you walk. It does not show that walking improves the prose you produce, and nothing we read measured dictated drafting on foot. For the full answer, see whether walking helps you write.

Dictating live into the manuscript and recording now to transcribe later suit different writers

There are two working patterns. In live dictation, you speak into the manuscript and see the words appear. In record-first dictation, you record sessions and transcribe them later. Each suits a different writer, and the choice affects privacy and sound quality.

Privacy depends on the tool. Apple's Mac Help says you can check whether general text Dictation is processed on your device and not sent to Siri servers ²⁵. Microsoft says Windows voice typing converts speech to text using online speech recognition ²⁶. An unpublished manuscript is a confidential document, so check where recognition runs before you dictate it. Our guide to voice privacy covers the questions to ask.

The edit pass is where a dictated draft becomes prose, so budget time for it

A raw dictated draft is not a finished text, and the errors are common. In a study of clinical notes dictated in 2016, the draft produced by speech recognition had 7.4 errors per 100 words ²⁷. After professional transcriptionists edited the drafts and physicians signed them, the error rate fell to 0.3 percent of words in the signed notes ²⁸.

These are clinical notes, not books, so the rates will differ for your text. The direction is the point. The draft is a first version, and the edit is where most of the accuracy comes from.

A workable edit pass runs in this order:

1. Read the chapter aloud, slowly, and mark every sentence that does not sound like you.
2. Search for every name in the names list, and check the spelling against your notes.
3. Check every opening and closing quotation mark, and every new paragraph in dialogue.
4. Fix the marked sentences with the keyboard, not by dictating again.
5. Read the chapter once more on screen, and mark anything left for the next session.

Software for authors needs custom words, long sessions and a good microphone

The software question has three parts: the tool, the words it can learn, and the microphone and voice that carry it through long sessions. Start with the tools you already have.

Built-in tools come first

Dictation is built into macOS, Windows has voice typing, and Microsoft 365 and Google Docs both offer dictation in their own ways. Check the current settings, browser support and app requirements before you rely on any of them, because these features change between versions. The built-in tools are often enough for a first book, and they cost nothing extra.

Paid tools make vendor claims that need checking

Dragon Professional is a paid option. Its vendor claims that version 16 lets employees dictate documents 3 times faster than typing, with up to 99 percent recognition accuracy ²⁹. That is a vendor claim, and we have not verified it. Subscription prices differ by reseller, so check the current price before you buy.

For a broader comparison, see our guide to dictation software. Yap, which we make, is an on-device dictation app for Mac and Windows, so it can be part of the same workflow. Current platforms and downloads are on our download page.

Microphones and session length affect the result

Apple advises that a headset microphone may help when Dictation is used in a noisy environment or a room with excessive echo ³⁰. Choose a quiet room for long sessions, and set up the microphone before the first scene.

Long sessions can strain the voice. One study of voice quality after dictating with speech recognition software found that one of the most consistent changes was a rise in pitch, which the authors read as the voice compensating for early voice fatigue ³¹. Plan breaks into the schedule, and stop a session before your voice starts to strain. For a microphone that suits long sessions, see our glossary entry on noise-canceling microphones.

The questions this guide answers

• What is the best dictation software for writing a book? The answer depends on whether your text can leave the device and whether you need custom words across many sessions.
• How many words per hour can you dictate? Speaking speed on a phone is far higher than composing speed by speech with earlier software, and planning sets the drafting pace.
• Can you dictate a novel? Yes, with an outline, a method for dialogue and names, and a serious edit pass.
• Does walking while dictating help you write? One 2014 study found that walking helps idea generation, and it did not test drafting.

For other kinds of dictation work, our guides on dictating emails, speech to text for students and dictation for journalists and content creators cover the same tools in other settings. For the basics of speed, see our guide to typing speed.