Wispr Flow Setup Guide
I have dictated over 1,05,400 words using Wispr Flow. This is the setup I wish I had on day one: Transforms, custom shortcuts, and the habits that actually stuck.
Step 1: Install Wispr Flow
Download it from wisprflow.ai. It is a Mac-first app, and the reason it works is that it is not locked to one app. It runs system-wide. Claude, Cursor, Notion, Gmail, Slack, your terminal. You hold the shortcut anywhere and it listens.
- Open the .dmg, drag it to Applications. Takes about 30 seconds.
- It will ask for Accessibility and Microphone permissions. Grant both. Without them, the system-wide part does not work.
- Sign in or create an account. Your Transforms and Insights sync to your profile across devices.
- Once it is running, look for the small microphone icon in your menu bar. That is where everything lives.
Step 2: Your First Dictation
Click into any text field. Hold Right Option (⌥) and speak. Release the key and it inserts. That is the whole interaction. It sounds too simple until you actually do it in Claude at 11pm when you are too tired to type properly.
- If Right Option clashes with another tool you use, go to Settings → Shortcut and change it. I left mine as default and it has never caused a conflict.
- Auto-edit is on by default. It quietly strips filler words and smooths out grammar before the text lands. Most of the time this is exactly what you want.
- One exception: vibe coding prompts with exact variable names or file paths. For those, glance at the transcription before you hit send. Takes one second.
Step 3: Transforms. The Feature That Changes Everything
This is the one most people miss. Transforms let you select any text in any app and run an AI instruction on it with a single shortcut, without opening a new tab, switching tools, or breaking your flow. You select, you press a key, and the text rewrites itself in place.
I use this probably fifteen times a day. To get there: click the Wispr Flow menu bar icon → Transforms.
The Two That Come Built In
Wispr Flow gives you two Transforms out of the box. Both are genuinely useful:
Improves clarity and conciseness. Removes redundancy, tightens sentences, keeps your voice intact.
Constructs an optimal AI prompt from your rough instruction. Great for improving vague prompts before sending.
How to Run a Transform
Four steps. Literally.
- Write or dictate something in any app.
- Select the text you want to change.
- Hit the shortcut: ⌥ + 1 for Polish, ⌥ + 2 for Prompt Engineer.
- Watch it rewrite in place. You never left the app.
Live example: Polish Transform
Wispr Flow highlights additions in green and removes redundant words with strikethrough.
Step 4: Build Your Own Transform
The two built-in ones are good. But the ones you build yourself are what make this feel like your tool. Any instruction you would normally type into Claude, "make this more casual", "turn this into a LinkedIn post", "structure this as a vibe coding brief", can become a one-key shortcut.
- Menu bar icon → Transforms.
- Hit + New Transform.
- Name it something obvious: "Casual", "Brief", "Hook". You will be pressing a shortcut in the dark at 8am.
- Write the instruction. Treat it like a prompt you would give Claude. The more specific, the better the output.
- Assign a shortcut. I use ⌥ + 3, ⌥ + 4, ⌥ + 5 for my customs.
- Save. It is live system-wide immediately.
Example Custom Transform Instructions
Things I Learned the Hard Way
- Say the whole thought, not just the gist. When you type, you trim. When you speak, do not. The auto-edit handles the rough edges. Your job is to say everything, including the context you would normally cut to save time. That context is what makes the AI output actually useful.
- Polish your AI prompts before you send them. Dictate the rough instruction, select it, hit ⌥ + 2 for Prompt Engineer. It turns "make this shorter" into a proper structured prompt. The output from Claude is noticeably better.
- Build the shortcut for the thing you do most. Not the thing you think you will do most. The thing you are already doing ten times a day. For me that was AI prompts. For a copywriter it might be email tone. Pick one and build it first.
- Vibe coding prompts need a second look. Auto-edit is your friend for prose. For prompts with component names, exact file paths, or logic chains, read the output before you hit enter. You are not checking grammar. You are checking whether "authentication middleware" became "authentication middleware" or something close but wrong.
- Look at Insights once a week. WPM, streak, peak time, top apps. I found out my most productive voice session is Friday at 9am. That is not what I expected. Knowing it means I plan accordingly.
- Friday morning is for voice. Planning, wiring automations, writing briefs. These all dictate better than they type. Block 30 minutes on a Friday morning and just speak your week's setup into your tools. You will feel the difference by noon.
This guide pairs with the blog post Voice Dictation for Vibe Coding: Why Speaking Beats Typing. Read it for the context on why this workflow matters.
Quick Reference: Shortcut Cheat Sheet
| Action | Default Shortcut | Notes |
|---|---|---|
| Start/stop dictation | Hold ⌥ Right | Hold to record, release to insert |
| Polish Transform | ⌥ + 1 | Select text first |
| Prompt Engineer Transform | ⌥ + 2 | Select text first |
| Custom Transform 1 | ⌥ + 3 | Assign in Transforms settings |
| Open Wispr Flow menu | Menu bar icon | Access settings, Transforms, Insights |
Harshal Saraf
Creative Director + AI Workflow Consultant
Harshal Saraf is a Creative Director and AI Workflow Consultant based in Indore, India. Under his practice ByHarshal, he sets up AI workflows for founders, agencies, and brands across India. He has dictated over 1,05,400 words using Wispr Flow, with 60% going directly into AI workflows and vibe coding sessions.