Voice input

Stop typing and start talking. Most AI apps already have a mic button: Claude desktop and mobile, ChatGPT, and /voice in the Claude Code CLI on Mac. Tap it and explain your intent like you are briefing a teammate at the next desk.

Two things happen. First, speed: people speak around 150 words per minute versus 40 to 80 typed, so you get roughly two to three times more context per minute (Stanford research puts spoken input near 3x faster for English). Second, and more important, behavior changes. When you type you compress and abbreviate. When you speak you finish the thought, so you naturally include the "why," the constraints, the edge cases, and the "one more thing." That is the real fix for short-prompt, weak-output cycles. It also lowers the friction of starting a prompt at all. Tribe.ai reported that voice-based context sharing far outperformed typing across their internal measures, though that figure is their own.

Reach for it whenever you are about to type more than a sentence: kickoff prompts, briefing an agent before a long task, describing a bug, explaining architecture, async hand-offs, or setting context at the start of a session. It is ideal on mobile or while walking and commuting.

Skip it when you need exact syntax like code, file paths, or URLs, and when you are in a noisy room, since accuracy drops. A good dictation layer can auto-tag files to soften the syntax weakness. One privacy note: Claude Code streams audio to Anthropic's servers to transcribe, while Claude mobile deletes the audio and does not train on it, which matters for sensitive work.