Prompt Flow: Voice Dictation Optimizer for Claude, ChatGPT, Gemini

About

Prompt Flow turns messy voice dictations into clean, effective prompts for Claude, ChatGPT, Gemini, and Meta AI.

Built for Wispr Flow users - but works with any voice dictation tool, or with typed input.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

THE PROBLEM

Voice to text is fast. Getting the LLM to understand what you actually meant is not.

Voice dictation has unlocked a 10x faster way to interact with AI. But raw transcripts are messy: filler words, false starts, vague phrasing, no structure. You either send the rambling version (and get rambling back) or stop to clean it up manually (which kills the speed advantage).

THE FIX

Press ⌘⇧Space (Mac) or Ctrl+Shift+Space (Windows) after dictating. Prompt Flow - a prompt optimiser and Chrome extension - cleans up the text in place and replaces it in the chat input. No popup, no copy-paste, no friction.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

CORE PHILOSOPHY

Most "prompt enhancers" make every prompt longer. They add scaffolding, role assignments, and bullet lists whether you wanted them or not.

Prompt Flow does the opposite. It matches the level of intervention to your input:

- A short, specific dictation gets light cleanup
- A rambling dictation gets compressed to its essential structure
- A vague dictation can be expanded with role + constraints + format - but only when you ask

The optimizer never invents domain frameworks, checklists, or considerations you didn't mention. Open-ended questions stay open-ended.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

THREE OPTIONS, THREE USE CASES

↓ TRIM TOKENS (default ON)
Forces output to be shorter than input. Removes filler ("um", "like", "you know"), hedging ("I think maybe"), and padding. Preserves every fact and every ask.

🔒 PRESERVE INTENT (default OFF)
Zero interpretive additions. Grammar and filler cleanup only. Output reads like a tidied transcript. Use this when you want exactly what you said, just cleaner.

＋ ADD STRUCTURE (default OFF)
Full scaffolding for vague prompts: role assignment, constraints, format specification, and model-specific techniques. Use when you have a vague idea and want a well-shaped prompt.

The default combination (trim tokens on, others off) handles ~80% of dictations correctly.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

MODEL-AWARE OPTIMIZATION

Prompt Flow detects which AI you're using and applies model-specific best practices from each provider's official documentation:

- Claude — XML tags for multi-section prompts, literal instruction phrasing
- ChatGPT — Imperative verbs, explicit format specifications
- Gemini — Numbered requirements, direct concise structure
- Meta AI — Conversational imperatives, clear length specs

The same dictation produces different optimized prompts depending on which AI will receive it.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

CONTEXT AWARENESS (optional)

Off by default. When enabled in Settings, Prompt Flow reads the last 3–4 messages from your current chat to resolve pronouns and maintain thread continuity.

So when you dictate "also add error handling to that", Prompt Flow knows what "that" refers to.

Chat content is sent only to your selected API provider. Never stored by Prompt Flow.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

BRING YOUR OWN API KEY

Prompt Flow doesn't sell you a subscription. You bring your own API key - free options are available from Groq and Google.

- Groq (Llama 3.3 70B) - FREE, fastest, recommended for most users
- Gemini 2.0 Flash - FREE, generous limits
- Claude Haiku 4.5 - Paid, highest quality output

Get a free Groq key in under 60 seconds at console.groq.com.

━━━━━━━━━━━━━━━━━━━━━━━━━━━

PRIVACY-FIRST

- API keys stored locally on your device using Chrome's secure storage
- No telemetry, no tracking, no analytics
- No external servers - Prompt Flow talks directly to your chosen API provider
- Chat context (when you enable it) is sent only to your provider, never to us
- Prompt history stored locally on your device, can be cleared anytime

━━━━━━━━━━━━━━━━━━━━━━━━━━━

SUPPORTED SITES

- claude.ai
- chat.openai.com / chatgpt.com
- gemini.google.com
- meta.ai

━━━━━━━━━━━━━━━━━━━━━━━━━━━

WHO IT'S FOR

✓ Wispr Flow users who want their voice dictations to land clean
✓ Anyone who finds typing prompts slower than thinking them
✓ Power users and prompt engineering enthusiasts who want one keyboard shortcut, not a popup workflow
✓ Privacy-conscious users who want bring-your-own-key, no subscription

━━━━━━━━━━━━━━━━━━━━━━━━━━━

GETTING STARTED

1. Install Prompt Flow
2. Click the extension icon and add a free Groq API key
3. Open Claude / ChatGPT / Gemini / Meta AI
4. Dictate (or type) a prompt
5. Press ⌘⇧Space (Mac) or Ctrl+Shift+Space (Windows)
6. Watch your messy text become a sharp prompt

That's it. No tutorial needed.