Long prompts get better answers than short ones, and voice input keeps the prompt rich without a heavy typing session.
Start voice input
On Mac press the Fn key twice inside the target field.
On Windows press Windows plus H to open voice typing.
On Chromebook enable dictation in Accessibility settings.
Click into the chat prompts before you speak so the caret is set.
Prompts vs code comments vs PR summaries
Prompts carry the ask. Code comments carry the intent. PR summaries carry the shipped change. Speak in each shape.
Draft in Zahvox for careful copy
For anything sent to Copilot Chat or attached to a pull request others will review, dictate the raw thought in Zahvox, tighten the wording, then paste it back. That extra pass keeps the tone steady and removes filler before the words reach a live surface.
Small habits that keep it readable
Say punctuation out loud, including commas and periods.
Keep one thought per sentence so cleanup is fast.
Do a short review pass at the end of every session.
Save common phrases in a snippet tool so voice input can call them.
Trust the second pass more than the first
The first pass from dictate prompts and follow ups in github copilot chat is a raw material, not a product. All the real quality comes from the edit that follows — trimming filler, fixing order, cutting the sentence that added nothing. Once you accept that split, the whole workflow gets calmer. You stop trying to speak perfect prose and you stop apologising for messy transcripts, because you know the second pass is where the writing actually happens.
Where Zahvox fits after the raw capture
Zahvox is the careful second pass that sits after the raw capture. The browser, the OS, or the app you used for dictate prompts and follow ups in github copilot chat gave you words on a page. Zahvox at /tool is where you paste those words, tighten the wording, fix the shape of the sentences, and copy the result back into wherever it needs to live. Raw capture anywhere, careful edit in Zahvox — that is the whole loop, and it stays the same whether the source was a laptop, a phone, or a meeting transcript.
