Using Voice for AI Dictation & Documentation
Speaking is 3x faster than typing. Learn how to use voice input with AI tools for faster documentation, meeting notes, and technical writing.
Why Voice for Documentation?
Documentation is the task every developer avoids. It's slow, tedious, and always falls behind the code. Dave Killeen's approach: speak your documentation instead of typing it, then use AI to clean it up.
The numbers are compelling: average typing speed is 40 WPM. Average speaking speed is 130 WPM. That's a 3x speed increase for the first draft — and AI handles the editing.
The workflow works for technical documentation, meeting notes, design decision records, PR descriptions, and any text that starts as spoken thought. Most developers think in spoken language before translating to written text anyway — why not skip the translation step?
The Voice Workflow
The three-step voice documentation workflow.
Speak your first draft
Use any voice input tool (macOS Dictation, Google Voice, Whisper) to speak your documentation. Don't worry about structure or polish — just get the content out. Speak as if you're explaining to a colleague.
AI cleanup and structuring
Paste the transcription into Claude or ChatGPT: 'Clean up this voice transcription into structured documentation. Add headings, fix grammar, organize into logical sections. Keep my voice and intent.' The AI transforms rambling speech into clean documentation.
Quick review and publish
Review the AI-cleaned version. It typically needs 2-3 minutes of adjustment — much less than writing from scratch. The key: your unique knowledge and voice are preserved, just organized better.
Tips for Effective Voice Input
Narrate Context Before Details
Start by explaining WHAT you're documenting and WHO it's for. 'This is a setup guide for new developers joining the backend team.' This context helps the AI structure the document correctly.
Use Verbal Markers
Say 'new section' or 'bullet point' to give the AI structural hints. 'New section: Authentication. Bullet point: We use JWT tokens. Bullet point: Tokens expire after 24 hours.' This produces better-structured output.
Don't Self-Edit While Speaking
The biggest productivity killer is stopping to correct yourself. Just keep speaking. The AI handles cleanup. Stopping to fix a sentence costs 10x more time than letting the AI fix it later.
Record Design Decisions as They Happen
When your team makes an architecture decision in a meeting, immediately speak the ADR: 'We decided to use PostgreSQL because...' Capture decisions in the moment when context is freshest.
