Talk to it: voice input
Build-along · ~8 min at your own keyboard
Most people talk faster than they type, so dictation is a cheap upgrade. Messy, unplanned input is fine — structuring it is the AI's job. Voice is one way in; a quick typed brain-dump works the same way if you can't or don't want to speak.
The long version: Voice input for Claude ↗
Mac: Mac: System Settings > Keyboard > Dictation. Press the dictation key twice (or the shortcut shown) while the cursor is in Claude.
Windows: Windows: press Win+H in PowerShell, Windows Terminal, or VS Code to dictate. Or install a Whisper app like Wispr Flow.
Do this
0/4Checkpoints
- ▸You should now see your spoken words appear as text in the input box.
- ▸You should now see a structured version with situation, what you need, and open questions.
- ▸You should now see every concrete detail from your input preserved inside the structure, with nothing added.
The three ways this goes wrong
- Dictation does not start, or nothing appears in the terminal. — Check that dictation is on and your microphone is allowed for the terminal or VS Code. Test it in a notes app first.
- Dictation inserts text but cuts off or mangles names. — Speak in shorter bursts and fix names by typing. Claude will usually infer the right name from your files.
- The structured version dropped a detail you said, or added one you didn't. — Compare it line by line with your input. Point out what's missing or invented and ask it to fix that. Faithful beats tidy.
Share what happened
Paste a few lines of your raw dictation plus the structured version Claude produced.
Strip API keys, passwords, and tokens before you paste. Swap them for [KEY].
🔒 Don't paste confidential, personal-health, or payment info.