Guides
Dictating AI prompts: the length is the point
The single biggest improvement most people can make to what an AI gives back is to say more about what they want. Typing discourages that. Speaking does not.
Last updated
Why this is different from ordinary dictation
Everywhere else, dictation saves time. With an AI assistant it changes the output, because the model can only work with what you told it.
You know the constraints. You know what you already tried, what the thing is for, who will read it, and what would make the answer useless. You type none of it, because typing it takes a minute and you would rather see what happens. So you get a generic answer, refine it twice, and spend three minutes instead of one.
Typed, nine words
write a follow up email to a client who went quiet
Spoken, twelve seconds
Write a follow-up email to a client who has gone quiet. Context: they asked for a proposal three weeks ago, I sent it, and nothing since. We've worked together twice before and the relationship is good, so this shouldn't read like a chase. I want one short paragraph that gives them an easy way to say "not now" without it being awkward. No exclamation marks, no "just circling back", and don't ask if they saw my last email — obviously they did. Sign off as Nasir.
Twelve seconds of speaking, and the first answer is usable. That is the entire argument, and it applies to ChatGPT, Claude, Gemini, Cursor and everything else with a prompt box.
The workflow
- Click into the prompt box In the browser or the desktop app — both are ordinary text fields.
- Latch the microphone rather than holding it Double-tap to keep it open. A good prompt is sixty to a hundred words and holding a key that long is unpleasant.
- Say five things What you want, the context, the constraints, what you have already tried, and what a bad answer would look like. That last one is the most under-used.
- Stop, read it A misheard name sends the model somewhere useless.
- Send
The five things worth saying out loud
- What you want. The obvious part, and the only part most people type.
- Context. Who it is for, what happened before, what the situation is.
- Constraints. Length, tone, things to avoid, formats you cannot use.
- What you already tried. Stops it suggesting the thing that failed.
- What a bad answer looks like. Astonishingly effective, and almost nobody does it.
Voice mode is not the same thing
ChatGPT and others have their own voice modes. They are conversational: you speak, it speaks back, and the interaction is a dialogue. That is a different product experience and good for some things.
Dictating a prompt is not that. You are producing text, which you can read, edit and refine before the model ever sees it. You stay in the normal interface with the normal history. For anything where the prompt matters, this is the better tool — and it works the same way in every AI product rather than only the ones with a voice feature.
| Built-in voice mode | Dictating the prompt | |
|---|---|---|
| Output | A spoken conversation | Text you can edit before sending |
| Where it works | Apps that have the feature | Every prompt box on your Mac |
| Audio goes to | The AI provider | Nowhere, with an on-device app |
| Good for | Thinking out loud, hands-free | Prompts you want to get right |
| Editable before sending | No | Yes |
The privacy point
Prompts contain more than people realise: customer names, internal systems, salary figures, medical details, half-formed business plans. The prompt goes to the AI provider, because that is what you asked for.
With a cloud dictation app, the audio also goes to a second company — the transcription vendor — which was probably not part of anyone's assessment. On-device recognition removes that hop and leaves only the transfer you intended.
Habits that make this stick
- Teach it your vocabulary. Product names, colleagues, technical terms. A prompt with three wrong nouns in it produces an answer about something else.
- Make a snippet for your standing context. "I'm a solicitor in England and Wales; assume English law; do not give US examples." Say three words, get the paragraph. See voice snippets.
- Dictate the follow-up too. Refinements are where people go shortest — "make it shorter" instead of explaining what was wrong.
- Speak the negative constraints. "Don't use bullet points, don't hedge, don't tell me to consult a professional."
- Keep going past the point it feels complete. Spoken prompts feel long at forty words and are usually better at eighty.
Questions
Is dictating a prompt better than ChatGPT's voice mode?
For prompts that matter, yes, because you get editable text and can fix a misheard name before the model sees it. Voice mode is better for hands-free conversation. They are different tools.
Does dictating make prompts longer?
Substantially, and that is the benefit rather than a side effect. People type nine-word prompts because typing sixty words is tedious. Speaking sixty words takes twelve seconds, and the answer is noticeably better.
Does my prompt go anywhere extra when I dictate it?
With a cloud dictation app, the audio goes to that vendor before the text reaches the AI provider. With an on-device app like VV it does not — only the prompt you sent reaches the AI.
Does this work with Claude, Gemini and Copilot?
With all of them, and in any browser, because a system-wide app types into whatever field your cursor is in. There is nothing provider-specific about it.