Where dictation is socially impossible
Open-plan offices, shared desks, meeting rooms with thin walls, trains full of commuters.
Dictation is quietly excellent when you are alone and socially impossible when you are not, and no software addresses that. Saying a client's name aloud, or reading out a salary figure, or dictating a message about the person opposite you, is a different act from typing it silently.
That constraint decides more than any feature comparison. It is also why the people who get the most from dictation at work tend to be the ones who move around: site visits, travel, driving between appointments, working from home, walking between buildings.
Two practical adjustments help. Dictate the harmless parts and type the rest, a trick that only works when both live in the same keyboard and not in separate apps. And keep dictations short, because a sentence sounds like half a phone call and a paragraph announces itself to the room.
The question your organisation will ask
Not "is dictation allowed". It is "where does the audio go".
Most organisations are relaxed about tools and strict about data leaving a controlled environment. Client material, personal data about colleagues, medical or legal detail, anything covered by a confidentiality agreement or a contract with a customer.
Cloud dictation means audio is transmitted to a third party for processing. Nothing sinister or unusual about that, and plenty of employers approve it. But it is a question that has to be answered by somebody, usually with a data processing agreement and a supplier assessment, and on a personal phone used for work nobody has done that assessment at all.
On-device recognition does not raise the question. Nothing is transmitted, so there is no processor to name, no agreement to check and no cross-border transfer to reason about.
We are not lawyers and this page is not advice. If your organisation has rules, they are the authority, not us.
What to check before relying on it
Whose phone is it. A managed corporate device may restrict which keyboards can be installed at all, and that decision belongs to whoever manages it.
What your policy says about voice. Some policies name dictation explicitly. Many do not, and the honest answer is that the person who wrote it did not think about it.
Whether the app transmits. The airplane mode test settles this in ten seconds and is the single most useful thing you can demonstrate to somebody asking.
What happens in password fields. On a work phone this matters more, because you are logging into more things.
Where dictation genuinely earns its place at work
Once those are settled, the value is concentrated in specific moments.
Between things. The five minutes after a meeting, walking to the next one, while the taxi is arriving. That is when the follow-up email either gets written or gets postponed until Thursday.
Long replies on a phone. The message that needs three paragraphs and therefore waits for a laptop. The backlog in your inbox is made almost entirely of those.
Notes immediately after a conversation. Detail decays fast. Capturing it in the corridor rather than at your desk two hours later is the difference between accurate notes and approximate ones.
Site and field work. Anyone whose job happens away from a desk gets the most out of this by a wide margin, because the alternative is not typing but remembering.
The vocabulary problem is worse at work
Work writing is dense with exactly the words recognisers fail on: client names, colleagues' surnames, product names, internal systems, acronyms nobody outside the building has heard.
That makes the personal dictionary less optional than it is for personal use. Ten minutes adding your accounts, your team and your recurring jargon converts dictation from frustrating to usable, and in LocalType those additions stay on the handset and are given to the speech model as context so the names start arriving correctly.
One caution specific to work: think before adding anything that is itself sensitive. A client list in your keyboard dictionary is still a client list.
Turn off the radios and watch
The recognition runs on the phone, from a model file no other app can open, so the audio is not transmitted and there is no third party in the process at all. That answers the question your organisation will ask, and the answer comes out of the design itself. It is not a promise about how we behave.
You can demonstrate it. Nobody has to take our word for it. Turn off mobile data and Wi-Fi, dictate, watch it work. If audio needed to reach a server, it could not.
Practically it is a keyboard, so it works in whichever mail, messaging or task app your organisation uses, without that app needing to support anything. Switching between speaking and typing happens inside the same message. The open-plan problem becomes manageable on that alone.
The rest follows from having no server: nothing to register for, no advertising, nothing inside the product recording what you do with it, no archive of your recordings, and backups that skip the app on purpose. Over a password field the microphone declines to appear. The network is contacted once, to collect the model you chose.
Three sizes, at 60, 190 and 539 MB, set speed against accuracy. Work dictation tends to favour the larger end, since a misheard client name costs more than a two-second wait. On the 4 GB test handset the middle size needed 4.1 seconds for 6.9 seconds of speech. Eighteen languages are covered by one multilingual model, identified from the recording, which matters in any organisation that operates across borders.
Personal phone, work messages
If you answer work email on your own handset, your organisation has usually approved the mail app and thought about nothing else. The keyboard you type those messages with was never assessed by anybody, because keyboards are invisible until somebody points at them.
None of that is an argument for panic. It is an argument for knowing the answer before you are asked, and the answer sits in the app itself, not in a policy document. A keyboard that never contacts anything cannot be the route by which something leaves, and you can establish that yourself in under a minute.
It also cuts the other way. If your dictation does transmit and your employer later decides that matters, the messages have already been sent. Checking is cheap and retroactive fixes are not.
What it actually replaces
Dictation will not replace typing at work, because a good deal of work writing happens in rooms where speaking is not an option.
What it replaces is the writing that was not happening at all: the follow-up postponed to Thursday, the note you meant to make, the reply that needed paragraphs. Those are the messages that cost you a day each and never appear on any list of things you did not do.