Does Dictation Change How You Write?

Dictation changes the draft less than most writers expect. In a controlled study, planning takes most of the time spent composing, whatever the method of output ¹, and novice dictators can learn within a few hours to match their handwritten speed and quality ². What dictation probably changes is the texture of the text. Spoken habits such as fillers, restarts and loose sentences may come along unless you remove them, so a dictated draft usually reads better after you have revised it by eye.

This guide reads the primary studies behind that answer, from a 1978 letter-writing experiment ² to a 2002 study of students with learning disabilities ³. It separates what they show from what forum posts and vendor pages claim, and it says where the evidence runs out. Yap, which we make, is an on-device dictation app. Nothing here depends on it: the revision steps at the end work with any dictation tool, built-in or third-party.

Composing is the core skill, and speaking or typing is secondary

The best-controlled evidence treats the output method as a minor variable. In 1978, John Gould and Stephen Boies, then at IBM, timed and rated one-page business letters that writers composed by writing, by dictating to a recorder and by speaking. Their headline result was that novice dictators can learn in a few hours to dictate with the speed and quality with which they write ². The same study found that planning took about two-thirds of composition time, regardless of the method ¹. The slow part of writing is deciding what to say, not the physical act of producing words.

Years of practice still help. Dictators with years of experience were from 0 to 25 percent faster than novices, depending on how complex the letter was ⁴. The novices also underrated themselves. They did not think they performed as well as the results showed ², and that confidence gap may be one reason first-time dictation feels worse than it is. For the habit itself, see how to get used to dictation.

Cognitive research explains why the method matters less. In Kellogg's 2001 experiments, college students wrote narrative, descriptive and persuasive texts, by longhand or on a word processor, while they detected probes asking what they were doing at that moment. The author concluded that planning, translating ideas into sentences and reviewing share one general-purpose working memory resource ⁵. Speaking does not remove that competition. You still plan, phrase and check, and only the way the words come out changes. That is why oral composition and voice writing are better understood as a change of medium than of skill.

The table lists the four studies that compare ways of producing text.

Two limits apply to the 1978 and 1983 work. The dictation went to a recorder or to a simulated listening typewriter ⁶, so it measures the composing skill rather than the errors of modern recognition. And the letters were one page long ¹. Longer documents, such as theses and books, have not been measured the same way, so the result should not be stretched to them without caution.

Dictated drafts are not reliably longer than typed ones

The common claim is that dictation makes writing longer, and the controlled evidence does not support it as a general rule. In Snider's 2002 study, essay length did not differ significantly between speech recognition, a human transcriber and the keyboard ⁷, even though quality and rate did ³. We found no controlled study showing that adults write longer when they dictate. Length probably depends on the writer, the task and whether they edit as they go, so a promise of longer drafts is not something the research backs.

The likelier effect is on register rather than length. Spoken language has its own patterns, which the section on spoken language habits below sets out. We did not find a study that measures those patterns in dictated adult prose, so treat that section as the evidence that the mechanism exists, not as a measurement of your drafts. If you write academic work, dictation for academic writing applies this to theses and papers.

Quality depends on revision, not on whether the draft was spoken

Quality is where the evidence is most favorable to dictation, and the reason is practice and revision rather than the act of speaking. Gould and Boies found that dictators learned to match the quality of handwritten letters within a few hours ². In Snider's study, students with learning disabilities produced higher-quality essays with speech recognition than with a keyboard ³. Neither result says a draft is finished. Both measure a draft at the point the writer stopped.

The cost sits in correction. Vertanen's 2006 paper opens by reporting that, when people dictate with a large vocabulary recognizer, correcting errors dominates task time ⁸. Correcting by speaking again is often misrecognized too, which costs time ⁹. Kraal and colleagues say novice users keep trying to fix errors by voice, which reflects a wrong mental model of the tool ¹⁰. Citing earlier work, they report that users who keep dictating for about 20 hours start to combine voice correction with keyboard and mouse correction ¹¹. The dictation mistakes guide covers the correction workflow in detail.

Feng and colleagues' longitudinal study points the same way. Users became more efficient with experience ¹², and the largest productivity gains came from better dictation quality and from using navigation commands ¹³. The skill that matters is editing in the medium, and it is learned. Dictation commands lists the commands for it.

Speaking removes the typing bottleneck, and the strongest evidence is for students with learning disabilities

Speech is far faster than typing for producing words. In Ruan and colleagues' 2018 phone study, English speech entry reached 153 words per minute, against 52 on a phone keyboard, and the speed counted the finished text, including correction time ^{14 15}. Composing is slower than either, because it includes planning, which the studies above show takes most of the time. Feng and colleagues cite earlier figures of 8 to 15 words per minute for composing with speech software ¹⁶, against normal speaking at 125 to 150 words per minute ¹⁷. Composing with the software therefore ran at about 5 to 12 percent of normal speaking speed ¹⁸. For the typing side of the comparison, see speaking vs typing speed.

The bars mix tasks, devices and years, so they show orders of magnitude rather than a head-to-head test. The gap between speaking and composing is the gap between producing a sentence and planning it, which is the point of the first section.

The same pattern shows up as time in clinical notes. In an observational analysis of Epic data covering 215,207 U.S. ambulatory physicians ¹⁹, physicians who mainly used transcription and dictation spent an average of 4.8 minutes per visit on notes ²⁰. In the adjusted comparison they spent 1.30 minutes less per visit than physicians who mainly used SmartTools ²¹. That is a time measure from one record system, and the authors caution that it does not show dictation causes shorter notes. It says nothing about note quality, but it is consistent with speaking saving time on a routine writing task.

For struggling writers, the best evidence is for students with learning disabilities. Snider's study found that speech recognition beat the keyboard on quality and rate ³. We found no controlled study that tests slow adult writers directly, so any benefit for them is an inference. For students, see speech to text for students with disabilities.

Talking through an idea is not the same as explaining it

Writers often recommend talking an idea through before writing it down. The research separates two activities. In a 2011 meta-analysis of 94 studies ²², Fox, Ericsson and Best found that the think-aloud effect, saying your thoughts as you work, was indistinguishable from zero, with a correlation of r = -.03 ²³. On average, plain thinking aloud neither helped nor distracted the people in those studies.

Explaining is different. Procedures that describe or explain thoughts and actions changed performance, raising it above silent controls ²⁴. All verbal reporting procedures tended to lengthen the task ²⁵. For a dictated draft, that suggests a practical split. Talking freely may cost little. Narrating your reasoning as you go may sharpen the thinking, but it takes longer. These were lab tasks rather than essays, so they show the mechanism, not the result for drafting.

For writer's block, the most direct evidence is about planning, not speech. Kellogg's 1988 experiments found that composing a rough first draft had no beneficial effect on quality ²⁶, while preparing a written outline led to higher-quality documents ²⁷. That study did not test spoken drafts, but its result suggests that a rough draft alone will not give a blocked writer a better piece, dictated or typed. If dictation helps, it probably helps by getting a plan out in words, and the outline result suggests that the plan itself matters. The practical options are in our question on writer's block.

Spoken language has different habits than written language

Spoken and written language differ in ways that can be measured. One marker is lexical density, the share of content words such as nouns, verbs, adjectives and adverbs in a text. Ure (1971), as summarized in a 2015 paper, placed the boundary at 40 percent, with speech below it and writing above it ²⁸. Eggins (1994), in a corpus analysis cited by the same paper, found 33 percent in speech and 42 percent in writing ²⁹. Both figures are secondhand and definitions of lexical density vary, but the direction is consistent. Our inference, which we have not measured in dictated prose, is that the first version of a dictated draft carries more of the spoken pattern, and that is the reason to edit it for density.

Speech also carries disfluency: the pauses, restarts and fillers that show a speaker is still working out what to say. In a corpus of task-oriented conversations, disfluency rates were higher when speakers discussed abstract figures, which the authors read as a sign of greater planning difficulty ³⁰. That study is about conversation, not dictation, but it describes what you hear when you dictate a hard passage. Software now removes some of it. Microsoft says Fluid dictation in Windows voice typing, on Copilot+ PCs, corrects grammar, punctuation and filler words as you speak ³¹. Filler removal changes the words that reach the page. Microsoft's description covers grammar, punctuation and fillers, not sentence structure, so you still have to revise for that. The difference between a verbatim record and clean text is covered in verbatim vs clean verbatim transcription.

Revising by eye turns a spoken draft into written prose

Errors are easy to miss by ear. In Hong and colleagues' 2018 study, people identifying speech recognition errors from audio output missed more than half of them ³². Adding a pause between words improved detection in online studies ³³. Read the draft on screen, where you can see every word, and keep listening for a final check.

Revision needs its own pass for a working-memory reason. Reviewing draws on the same pool as planning and translating ⁵, so planning and polishing in the same breath compete for the same space. Dictate the draft first, then revise it as a separate step.

1. Dictate in short paragraphs. Keep each pass to one idea, so a wrong sentence is easy to find and replace.
2. Read on screen. Find errors and roughness by reading, and save the audio check for the end.
3. Cut the spoken material. Remove fillers, restarts and openers such as "so basically".
4. Split the long sentences. Dictated sentences run long, so break them where you would pause.
5. Check names, numbers and terms by hand. A reader checks these first, so verify them yourself.

The correction research above suggests that the second pass is where most of the time goes, so budget for it. How to dictate covers the drafting technique that makes that pass shorter.

The studies settle the method question, but not long or original drafting

The evidence settles how much the output method matters for short, practiced writing. It does not settle long or original drafting. The table gives the strength of each claim.

Three things remain unknown. The research does not compare dictated and typed text over a whole book or thesis. It does not show whether spoken habits fade with years of dictation. And most controlled studies predate modern recognition and AI cleanup, so they cannot tell us how those tools change the register of a draft. Each of these would be worth a direct test before you rely on it.

To keep reading, start with how to dictate and dictation vs typing. For the questions this guide answers in short form, see is dictated writing worse than typed writing? and why does dictated text sound different from typed text?. The overview of the whole topic is dictation and voice typing.