What dictation is genuinely good at
The blank page problem is a speed problem in disguise. You know roughly what you want to say, and the gap between thinking it and getting it down is wide enough that you lose the thread and start editing before anything exists.
Speaking closes that gap. Most people talk at two or three times the speed they type on a phone, and considerably faster than they type carefully. That difference is enough to change how a first pass feels: you get the shape of a scene, an argument or a chapter out while you still have it in your head.
Dictation is particularly good for the material that arrives at inconvenient moments. Walking somewhere, standing in a queue, halfway through cooking. Those are the moments an idea shows up, and they are precisely the moments a laptop is not available.
What it is bad at, reliably
Knowing what it is bad at is what makes it usable. Not knowing is why people try dictation once and give up.
Sentences get longer. Spoken language runs on breath and momentum, not punctuation. Dictated prose tends towards long, comma-spliced sentences that read fine aloud and badly on a page.
Structure disappears. You cannot see what you have written while you are saying it, so paragraphs blur, you repeat a point you already made two minutes ago, and transitions go missing.
Names and technical terms come out wrong. A recogniser has never met your characters, your subject's surname, or the invented word your worldbuilding depends on. It substitutes something ordinary that sounds similar.
Punctuation is inferred, not dictated. Most modern systems work it out from your rhythm, and do not expect you to say "comma". Natural to use, and occasionally wrong in ways you have to catch.
None of that is a defect to be fixed by a better app. It is what happens when you produce language with your mouth instead of your hands.
A workflow that survives all of it
The writers who stick with dictation nearly all converge on the same three habits.
Separate the passes completely. Dictate without correcting anything. Do not stop to fix a wrong word, do not reread. The value is momentum, and every correction destroys it. Editing happens later, with your eyes, in a different sitting.
Speak in whole sentences, deliberately. Decide the sentence, then say it, then pause. Trailing off mid-thought gives the recogniser nothing to work with and produces exactly the mush you will hate later. This is a learnable habit and it takes about a week.
Mark the places you know are wrong. Say a distinctive nonsense word, the same one every time, where you want to come back. Your marker transcribes to something consistent, so you can search for it during the edit and skip the reread that hunts for gaps.
Teach it your vocabulary before you start
This is the single highest-return thing a writer can do, and most people never do it.
If your app lets you add words, put your recurring proper nouns in before you dictate anything long: character names, place names, the surnames of the people you are writing about, the jargon of your field. LocalType keeps those on the handset and hands them to the speech model as context, so they start arriving spelled correctly instead of being approximated.
Doing this once at the start saves more editing time than any other single change. You end up with a draft you can actually work with, and not one where every third proper noun needs replacing.
Why a keyboard beats a dictation app for this
Writers often reach for a dedicated transcription tool and end up back with the keyboard, for a specific reason.
A dictation app produces a block of text somewhere else, which you then move into your actual document. That is fine once. Over a long project it means constant copying, pasting and reconciling versions, and it makes the small additions, a line into an existing paragraph, a note in the margins of a draft, disproportionately annoying.
A keyboard puts the words where your cursor already is, inside whatever you write in. Notes app, document editor, email to yourself, the manuscript file. You are typing along, you switch to speaking for a paragraph, and you carry on typing. Nothing moves between applications.
You can also dictate straight into the middle of existing text. Most editing-stage additions land exactly there, in the body of something already written.
The offline part matters more than writers expect
Writing happens in odd places. Trains, aeroplanes, a café with hostile Wi-Fi, a cottage with one bar of signal, a garden with none.
Cloud dictation fails in exactly those environments, and it fails quietly. It waits, it times out, it loses the sentence you just delivered perfectly. There are few things more discouraging mid-flow.
LocalType runs its speech model on the phone, so the connection is not part of the process. Once the model is downloaded, dictation behaves the same in airplane mode as it does at home. Internet access is asked for exactly one purpose, fetching the model you chose, and never again.
There is a second reason this appeals to people who write for a living. Drafts are not finished work, and plenty of writers are uneasy about unpublished material, interview notes, or anything under embargo travelling to a server they do not control. With on-device recognition there is no upload, no account, and no archive of your recordings anywhere.
Choosing how much accuracy to buy
Longer dictation shifts the calculation towards a larger model.
Three sizes ship with LocalType: 60 MB for speed on limited hardware, 190 MB as the default for most phones, and 539 MB for the best handling of accents, fast speech and background noise. On a test phone with 4 GB of memory, the middle size turned 6.9 seconds of speech into text in 4.1 seconds, while the largest takes roughly twice as long as you spent speaking.
For quick replies, the wait on a larger model is irritating. For a fifteen-minute drafting session it is invisible, because you are talking rather than watching the screen, and every misheard word you avoid is a correction you do not make later.
Draft on the move, edit sitting still
One more habit to borrow from people who work this way: separate where you dictate from where you edit.
Drafting by voice suits movement. Walking, commuting, pacing a room while you think. Editing suits stillness and a screen. Trying to do both in one sitting tends to produce a session where you neither draft freely nor edit properly.
Because dictation goes directly into whatever app you already have open, the draft can land in a notes app, a document or an email to yourself, and be picked up at a desk later. Nothing has to be lifted out of a transcription window first, which is the friction that quietly kills this workflow when the dictation lives in a separate app.
A faster draft, and a messier one
The result is faster to produce and messier than one you typed. Longer sentences, weaker structure, some words wrong. All of it fixable, and none of it requiring you to have stared at a blank page.
If your bottleneck is getting words down at all, dictation removes it. If your bottleneck is knowing which words, it will not help you, and no tool will.