How to Talk to Your Computer (So It Does What You Mean)
Talking to a computer is not the same skill as talking to a person. A person fills in your gaps; a machine transcribes exactly what it hears. The people who get the most out of voice input follow one rule: pick the right register for the job. There are three.
Register one: dictate
Dictation is prose. Email drafts, documentation, commit messages, journal entries. You hold a hotkey, speak in complete clauses, and release. The transcript lands wherever your cursor is. The mistake most beginners make is speaking the way they think: fragmented, self-interrupting, full of "um". The engine writes all of it down. Speak the way a newsreader talks: one clause at a time, finished thoughts, no false starts.
Register two: command
Commands are not prose. They are verb plus object, stated plainly: "open the terminal", "summarize this page", "run the tests". When a command fails, the fix is almost never volume or speed. It is structure. Say what you want done and to what, in that order, and drop the politeness. Your computer does not mind.
DictatorFlow keeps this register on its own hotkey (ctrl+alt+j) so a command never ends up pasted into your editor as text. Separate triggers for separate registers is the whole trick: the desktop client gives dictation, commands, conversation, and deep research their own keys because mixing them is how you get a transcript where an action should have been.
Register three: converse
Conversation is for thinking out loud: brainstorming, asking follow-up questions, working through a problem while your hands are busy. Here fragments are fine, because the model responds and you iterate. This is the register where voice beats typing most clearly, since half of typing-based chat is formatting your thought into something presentable. Spoken, you just say it.
When a transcript comes out wrong, do not reach for the keyboard to patch one word. Re-speak the whole sentence. Speech engines score context across a phrase, so a clean second take usually fixes every word in it, while a hand-edit fixes one and leaves the rest broken.
Make the trigger automatic
Voice input fails for one reason more than any other: friction. If capturing a thought takes four clicks, you will stop capturing thoughts. A global hotkey that works in any app removes the decision. Press, speak, release, keep working. After about a week the keypress disappears from your awareness and voice becomes just another way text gets onto the screen, like paste but faster. Try the browser version first if you want to feel the loop before installing anything.