Answer-Box
Dictation tools like Raycast Dictation, VoiceInk, or superwhisper can boost your writing speed from 50 to 150 words per minute, but achieving reliable, ready-to-use output requires a structured approach: build custom vocabulary, set global instructions, and create app-specific styles. This method, refined over weeks of daily use, turns rambling thoughts into polished emails, terminal commands, or GitHub comments without tedious editing.
The Three Layers of Reliable Dictation
Most people give up on dictation because of recurring mistakes and missing context. The fix lies in three progressive layers: vocabulary, global instructions, and styles. Start by adding only the words that repeatedly fail—friends' names, technical terms, or abbreviations—rather than preloading every possible term, which can bloat the model. Next, set global instructions that apply to every transcription, such as always using lowercase or preserving code-switching between languages. Finally, create styles (or modes) that post-process the raw transcription for specific apps or tasks, making the output immediately usable.
Vocabulary: Fix Recurring Mistakes
Even advanced models struggle with personal vocabulary. When you notice the same word being misrecognized—like a colleague's name or a technical acronym—add it to the tool's custom vocabulary. Don't try to anticipate every term upfront; only add words after you spot a real, repeated error. This keeps the model focused and avoids unnecessary context that could degrade performance.
Global Instructions: Set Universal Rules
Global instructions steer every transcription without needing per-task configuration. For example, you can enforce lowercase output, define voice commands like "emoji …" to insert emojis, or set rules for handling incomplete sentences. These instructions act as a baseline that all subsequent styles build upon, ensuring consistency across different uses.
Styles: The Game-Changer for App-Specific Output
Styles are the most powerful feature for making dictation a daily habit. They take the basic transcription and format it for a specific destination—email, terminal command, GitHub comment, or note. Raycast Dictation can automatically apply a style based on the frontmost app, so dictating on github.com triggers a style that turns your rambling thoughts into a concise, friendly, and objective developer comment with proper code formatting. Similarly, a style for Ghostty (terminal) converts natural-language requests into paste-ready zsh commands, using macOS-compatible syntax and avoiding destructive actions.
Real-World Examples
One practical style transforms a dictated review into a GitHub or Linear comment. It preserves technical details, uses backticks for code terms (e.g., `useState`), and removes hedging while staying professional. Another style for the terminal turns "find all PDFs modified last week" into a single `fd` command. These styles eliminate the need to manually correct formatting, letting you focus on the content.
Tips for Daily Dictation
To make dictation reliable over time, dictate one idea at a time. This gives the style a focused task, limits the impact of transcription failures, and makes corrections easier. Assign a keyboard shortcut that feels automatic—many users map it to the Function key or a Hyperkey combination. Finally, spend 15–20 minutes each week reviewing your dictation history. Identify recurring mistakes and adjust the appropriate layer: microphone issues, missing vocabulary, or style prompts that need refinement. This "compound engineering" approach ensures each session improves the next.