Prompting an AI assistant well usually means writing more than a single line — constraints, an example, a format the answer should take, context the assistant would otherwise have to guess at. That is a real piece of writing, typed into a chat box dozens of times a day by anyone who has made an assistant part of their normal workflow, and it adds up the same way any other repeated typing does. This page is about replacing the typing, not the prompting itself: the app has no idea what a good prompt looks like and never touches the wording once it lands.
What actually slows down a day full of prompting
Rarely the thinking. Someone who prompts assistants regularly usually knows within a few seconds what they want to ask and how they want it phrased — the constraints, the tone, the example to include. What takes the time is typing that decision out in full, one keystroke at a time, dozens of times across a working day, on top of whatever else is already being typed that day.
Where a held key fits into an actual day of prompting
The prompt itself, said instead of typed
Hold ⌥ Space, say the prompt exactly as you'd want it read, let go — it lands in the assistant's text box, through the accessibility API where the tool supports it and a synthetic paste otherwise, clipboard restored either way.
Rewrite this function to handle the empty-array case explicitly, keep the same signature, and add a one-line comment explaining why the check is there.
Reading it back before it's sent
A prompt that's been typed by a held key sits in the box exactly as said, readable and editable before it's sent — an advantage over a live voice conversation with an assistant, where the words are already being acted on by the time a mistake is noticed. A misheard name, a wrong number, or a dropped constraint is a two-second fix in a text box and a much harder one once an assistant has started answering the wrong version of the question.
Follow-up prompts, dictated in a running thread
A long back-and-forth with an assistant is a series of short prompts more than one long one, and each follow-up dictates the same way the first prompt did — held key, said, released, landing wherever the cursor already is in the thread.
System prompts and instructions, drafted the same way
Writing a reusable instruction — for a custom assistant, a saved prompt template, or a project's system message — is prose like anything else on this page, and benefits from being said rather than typed, especially when several drafts are being tried in a row before one is kept.
What a day of prompting actually looks like, kept locally
- 09:155.8sRewrite this function to handle the empty-array case explicitly, keep the same signature.Claude
- 10:404.2sSummarize the three options above into a table with cost and effort as columns.ChatGPT
- 15:036.9sExplain why this query plan chose a sequential scan instead of the index, in plain language.Cursor
Every dictated prompt is written to local history the instant the key is released, before anything downstream happens to it, and that history stays on the machine it was said on. Nothing about a prompt is uploaded anywhere by this app to be processed, and it works the same with the network off — though the assistant on the other end of the prompt still needs its own connection to answer.
What this deliberately does not do
It does not talk to an assistant on your behalf, does not improve a prompt's wording, and has no memory of which assistant you prompt most or what a good prompt for it looks like. The full comparison against ChatGPT's voice mode covers the structural difference in more detail: one of these has a conversation and replies, the other only types what you said into whichever box has focus, and the two are not really competitors so much as two different tools that happen to share a search term.
It also has no custom vocabulary, so a project's specific terms, a model name, or a library the assistant needs to know about comes out however the engine hears it and needs the same quick check any dictated prompt would get before it's sent.
Where a rewriting tool does more than this one
Aqua Voice and Wispr Flow both lean further toward tidying spoken text before it lands — smoothing a rambling instruction into a cleaner one. That is a real capability neither this app has nor tries to: what lands in the prompt box here is exactly what was said, filler words and all, on the theory that a prompt's exact wording is often deliberate and worth keeping under your own control rather than a model's.
What happens if a browser tab reloads mid-prompt
- 1Key downthe target is chosen here
- 2You speaklevel meter, over your work
- 3Key up
- 4Written to historybefore the engine is asked
- 5Engineon your Mac
- 6Typedabout half a second
A chat interface that reloads, a browser tab that loses focus, or an assistant's own connection dropping mid-response are ordinary hazards of working in a browser-based tool, and any of them can interrupt a prompt that's mid-dictation. What changes is where the words already are: written to local history the instant the key is released, before this app's own transcription engine is even asked to produce a result — recoverable from history and pasted back into a fresh tab rather than re-said from memory.
Prompting across a desk and a laptop
A licence covers three devices, each installed separately with no account and no sync between them. Someone who prompts from a desktop most days and from a laptop occasionally gets an independent history on each machine — a prompt dictated at the desk does not appear on the laptop automatically, worth knowing before assuming a saved prompt will simply follow across machines.
A hotkey that survives a browser and an IDE's own shortcuts
Chat interfaces run in a browser, and coding assistants often run inside an IDE — both environments already claim dense sets of their own keyboard shortcuts. The default ⌥ Space is chosen to avoid the obvious collisions, but it's worth a quick check in System Settings, under Keyboard Shortcuts, against whichever tools are already installed. The general page on setting up a dictation hotkey covers that check in more detail.
Where this fits for a developer prompting a coding assistant specifically
The page on dictation for developers covers the wider case of prose in a developer's day — commit messages, tickets, PR descriptions — of which prompting a coding assistant's chat box is one part among several, using the same held-key mechanism throughout.
What it costs and what it runs on
$19 once, three devices, macOS 13 Ventura or later, Apple Silicon or Intel. The thirty-day trial runs the complete app, no card and no account, long enough to carry a real week of prompting across whichever assistants are already part of the workflow, and judge whether saying a prompt beats typing it once the habit sets in. The general page on Mac dictation covers the same mechanism applied to the rest of a working day beyond prompting specifically.
Why a long prompt benefits more than a short one
A one-line question types out fast enough by hand that dictation barely changes the arithmetic. The prompts that actually cost time to type are the longer ones — several constraints stacked together, a worked example included so the assistant knows the expected shape of an answer, a block of context copied in from somewhere else and then explained. Those are the prompts where saying the whole thing in one breath, rather than typing it sentence by constraint by example, saves the most real time across a day that includes a lot of them.
What changes when the same prompt gets reused across assistants
Prompting rarely stays confined to a single tool for very long. A question tried in one assistant that comes back vague gets tried again, sometimes worded slightly differently, in a second one to compare the answers. Because this app works the same way in any text field, moving a prompt from one assistant's chat box to another's costs nothing beyond re-dictating it or pasting it from history — there is no assistant-specific setup, no separate integration to configure, and no difference in how the words get typed depending on which assistant is about to read them.
Dictating context, not just the question
A prompt is often mostly context and only briefly a question — pasting in a stack trace, a paragraph from a document, or a snippet of code before asking what to do about it. The context itself is frequently typed rather than pasted, when it originates as something the person is describing rather than something already sitting in a file, and it dictates the same way the question that follows it does: as one continuous piece of speech, landing in the same box, rather than a separate step requiring a different tool.
Reading a prompt back before it's sent, one more time
It's worth restating this because it's the single most useful habit a dictated prompt asks for: a quick read before hitting send. A misheard word in a prompt doesn't just produce a typo the way it would in an email — it can send the assistant down an entirely wrong line of reasoning, and fixing that after the fact costs more than the ten seconds a read-back would have taken. The read-back doesn't slow prompting down in any way that matters; it's the same instinct anyone already has for a typed prompt, applied to a dictated one, and it's worth building into the habit from the very first prompt rather than after the first prompt that goes wrong.
Prompting from a document already open, without switching apps
A common pattern: reading a paragraph in one window and prompting an assistant about it in another, switching back and forth to compare the source against the question being asked of it. Because the dictation key works the same in any text field, the prompt can be said the moment the right wording comes to mind, without first switching to the assistant's window to start typing — the key can be pressed while still reading the source document, so long as the assistant's chat box is the field with focus when the key goes down. That small removal of a context switch is easy to undervalue until a day of comparing several documents against an assistant's answers is spent without it.
Who this actually suits
Someone who prompts an AI assistant dozens of times a day, wants their exact wording preserved rather than smoothed, and is prompting from a Mac rather than a phone, has a straightforward reason to try the thirty-day trial against a real week of prompting.
Someone who wants a tool that tidies a rambling prompt into a cleaner one automatically, or who wants to talk to an assistant conversationally rather than type a prompt into a box, is better served reading the comparison against ChatGPT's voice mode first — this app was built to type what's said, not to prompt anything on its own behalf.