How to Use Voice Dictation with AI: The Voice-First Workflow
How I write newsletters, emails, and AI prompts by talking instead of typing. The exact voice-first workflow, tools, and setup to write 3x faster.
I write most of my first drafts without typing a word. Newsletters, emails, Slack messages, instructions to my AI tools: I talk them. Most people talk around three times faster than they type, and if you're someone who thinks out loud, the keyboard has been your bottleneck the whole time.
Here's the full voice-first workflow: the setup, and the four ways I use it every day.
Step 1: Set up a dictation tool that works everywhere
The tool that runs my whole day is Wispr Flow. It sits on my computer with a shortcut key, works inside every app, and transcribes almost in real time. What makes it better than the dictation built into your other tools is that it learns you: it knows I keep texts casual and email a little more formal, it formats bullets and paragraphs when I'm brainstorming, and it keeps a dictionary of my clients' names so I'm not re-correcting them. (Not sponsored. It's just the tool I'd save first in a fire.)
Install it, learn the hotkey, and test it in three apps so you trust it everywhere: your AI chat, your email, your Slack.
Step 2: Talk your first drafts (the walk method)
When I have something to say, I go for a walk and talk the whole idea through. No outline, no structure, just the thought as it shows up. I record it into my notetaker, then hand the messy transcript to Claude:
Here's a voice transcript of me thinking through a piece. Clean it up,
give it a shape, and hand me back a short outline and a few opening
lines in my voice. Keep my phrasing wherever it's good. My stream of
consciousness never comes out in order, so re-sequence it so it flows.
The talking is the draft. By the time I sit down, the blank page is gone and all that's left is polishing.
One power move inside this: throw research requests in mid-thought. I'll say "find me the studies on how much faster people talk than type, pull that in here," and it drops in as a note. When I hand the transcript off, the research comes back done.
Step 3: Give your AI instructions out loud
I give better instructions when I talk. When I type, I under-explain, because typing is work. Talking takes the pressure off, so the side context comes out: the asides, the background, the where-I-want-this-to-land. That extra context is exactly what makes AI output good.
So when you brief your AI on a task, hit the voice icon and walk it through like you're training a new teammate: the steps, your expectations, what done looks like, what above-and-beyond looks like. Three to five minutes of talking beats a one-line typed prompt every single time.
Step 4: Review out loud (this one trains your AI)
When I review a draft, I keep dictation on the whole time and talk through my edits as I read. What's working, what's off, what sounds like me and what doesn't.
That does two jobs at once. The piece in front of me gets better, and every future piece gets better, because I'm giving my AI feedback instead of silent corrections. Say why something's wrong, then tell it to save the feedback to its rules. (The full system for that lives in the rules doc guide.)
Bonus use: talk to think
Sometimes the person I need to talk it out with is just me. After a hard meeting or a tangled decision, I record myself processing it. The thoughts get out of my head and land somewhere I can actually look at them, and my AI can turn the mess into an action plan. If you're a verbal processor and an overthinker, this is the ten minutes that stops the spiral.
FAQ
Is talking really faster than typing? For most people, yes, around three times faster. The bigger win isn't speed though. It's that talking removes the pressure to word things perfectly, so you give more context and better detail than you would by typing.
Do I need a special dictation app, or can I use my phone's built-in one? Built-in dictation works for short bursts. For a real voice-first workflow you want a tool that works across every app, formats as you speak, and learns your vocabulary. That's why I use Wispr Flow on desktop.
How do I turn a rambling voice memo into a usable draft? Hand the raw transcript to your AI with one instruction: clean it up, shape it, keep my phrasing where it's good, and re-order it so it flows. Rambling is fine. Re-sequencing is the AI's job, not yours.
What should I dictate first if I'm new to this? Your next three Slack messages or emails. Low stakes, instant payoff, and you'll feel the speed difference the first day.