Switching from Typing to Dictation for AI Workflows
Dylan Davisgo watch the original →
the gist
Dictating to AI models instead of typing increases output quality by encouraging users to provide more context and reduces the friction of self-editing.
The Case for Dictation
Moving from typing to dictation is a primary workflow optimization for AI power users. While typing speeds are typically limited to 40 words per minute and often involve real-time self-editing, dictation allows for speeds of 150 words per minute. The primary benefit is not just speed, but the quality of the AI response. Because users tend to ramble and provide more conversational context when speaking, the AI receives richer input, leading to more accurate and useful outputs.
Implementation and Best Practices
To maximize the effectiveness of dictation, users should distinguish between native dictation (speech-to-text) and advanced voice modes. Advanced voice modes are designed for conversational back-and-forth and often interrupt the user, whereas dictation simply converts speech to text for the model to process.
- Start with a goal: Explicitly state the objective at the beginning of the dictation, then provide the necessary context and background information.
- Ignore filler words: Do not waste time proofreading or removing filler words like "um" or "like." Modern LLMs are capable of interpreting intent despite these disfluencies.
- Use incremental chunks: For complex feedback or long tasks, dictate in 20 to 30-second segments rather than one long take to prevent degradation in dictation quality.
- Choose the right tool: Start with the native dictation features in ChatGPT or Claude. For users who want to dictate into any application, third-party tools like WhisperFlow or SuperWhisper are recommended because they use AI to clean up punctuation and typos in real-time.
Edge Cases for Typing
Typing remains necessary in specific scenarios where precision is required. This includes entering addresses or specific character strings where dictation might misinterpret numbers or formatting. Additionally, while desktop agents like Claude's Computer Use or OpenAI's Operator can often handle their own screenshots and error logs, users relying on browser-based interfaces may still need to type or paste error codes manually when troubleshooting.