Why email is the obvious candidate
Most phone writing is short. A reply, a message, a search. Email is the outlier: it is where you are expected to produce actual paragraphs, on the device least suited to producing them.
So the queue builds up. Messages you will answer properly later, from a laptop, which then becomes tomorrow, which becomes a backlog you dread.
Speaking removes the physical bottleneck entirely. Three paragraphs take about ninety seconds to say and roughly ten minutes to thumb. That difference is the whole reason to bother.
Dictated email sounds different, and usually better
The common worry is that dictation will make you sound unprofessional. In practice the opposite tends to happen, for a reason that is easy to miss.
Typed email drifts towards formality and padding, because typing is slow and people reach for stock phrases to fill the space. "Just following up on the below." "Please find attached." "I hope this email finds you well." Nobody speaks like that.
Dictated email comes out closer to how you would explain something to the person if they were in front of you: more direct, more concrete, and shorter. For most work correspondence that is a clear improvement and not a risk.
The genuine risk lies elsewhere.
Where it goes wrong
Length. Speaking is easy, so you say more than you would have written. A dictated email usually runs longer than it needed to, and the recipient notices.
Structure. You cannot see the message while producing it, so the paragraph breaks land badly, the ask arrives at the end instead of the top, and the whole thing reads as a monologue.
Names. Colleagues' surnames, client names, product names, acronyms. These are the words most likely to come out wrong and the most embarrassing when they do, because the person is reading their own name spelled incorrectly.
Punctuation. Modern systems infer it from your rhythm rather than expecting you to say it. Mostly right, occasionally not, and a misplaced full stop can change a sentence more than a misheard word does.
A routine that takes fifteen seconds
Say the ask first. Decide the one thing you want before you start speaking, and say it in the first sentence. This single habit fixes most of the structure problem, because the rest of the message then has something to organise itself around.
Speak in complete sentences and pause between them. That gives the recogniser the rhythm it uses to place punctuation, and it stops you producing one 90-word sentence.
Read it before sending. Not for typos, for length and for names. Cut the paragraph you added because talking was easy. Check every proper noun.
That last step is the one people skip, and it is the only one that is not optional. Email goes to real people and stays in their inbox.
Add the names you use, once
Since names are the main failure, deal with them directly. The alternative is correcting them forever.
If your keyboard lets you add vocabulary, put your colleagues, your clients, your products and your acronyms in once. LocalType stores those on the handset and passes them to the speech model as context, so the words start arriving spelled correctly instead of being approximated by whatever else sounds similar to them. They also improve the typing suggestions, since one app handles both.
Five minutes of setup removes the most common reason people abandon dictation for work correspondence.
Where the recognition happens is not a small detail here
Personal messages are one thing. Work email is routinely the most sensitive writing anybody does on a phone: salaries, contracts, client problems, medical appointments, things covered by an agreement you signed.
With cloud dictation, that audio is transmitted to a server to be turned into text. That is how the architecture works, and for many people and many employers it is entirely acceptable.
With LocalType it is not transmitted, because the speech model is a file on your phone doing the work there. There is no upload, no account attached to the keyboard, no advertising in the product and no behavioural tracking inside it. No archive of your recordings is kept, since the audio becomes text and is finished with. The app's data is deliberately excluded from Android's cloud backup, and the model sits in storage only the app can read.
One limit decides whether the rest is any use to you. Once the words are in your email client and you press send, they are governed by that client and your mail provider. The keyboard's part ends at the moment the text appears in the field, exactly as if you had typed it.
In password fields dictation switches itself off, which matters more than you would think on a phone where mail apps and login screens sit next to each other.
It also works when your email client does not
Email gets answered in awkward places: a train tunnel, a plane, a hotel with hostile Wi-Fi.
Because nothing is being requested from anywhere, dictation carries on in all of them. You compose the reply now and the mail client sends it when the connection returns. Internet access is requested for exactly one thing, fetching the speech model you chose, and never again afterwards.
Model size matters for this use. Longer messages favour accuracy over speed, so the 190 MB default or the 539 MB option make more sense than the 60 MB one unless your phone is short of memory. On a test phone with 4 GB of memory, the middle size turned 6.9 seconds of speech into text in 4.1 seconds.
Signatures, quoted replies and the mess underneath
The cursor is the problem here, not the recognition.
Phone email is full of text you did not write: a signature, the quoted message you are replying to, sometimes a legal footer. Dictation drops words wherever the cursor happens to be, and on a small screen the cursor is easy to misplace, so it is entirely possible to deliver two careful paragraphs into the middle of somebody else's quoted text.
Two habits prevent it. Tap into the body and check the cursor sits above your signature before you start speaking. And after a long dictation, scroll up before sending. The preview is not to be trusted.
This is the most common way a dictated email goes out looking careless, and it has nothing to do with recognition quality at all.
Four habits that do the work
Dictation makes email faster and slightly longer. Say the ask first, speak in sentences, add your colleagues' names once, and read it back before sending.
Do that and the backlog stops being a backlog, without any of it leaving your phone on the way.
One last suggestion. Try it first on the messages you are least worried about, the routine confirmations and short answers. Leave the difficult one you have been avoiding for a week until later. You will learn where it mishears you on writing that does not matter, and by the time you reach the difficult message the habit is already there.