Make Claude Write in Your Voice, Not a Robot's

With no instruction, Claude writes to the safe average of everything it has read - hedged, tidy, and faceless. That is not a limit of the model; it is missing input. Voice is learnable, but only if you show it what yours is. Adjectives like 'make it punchy' barely move it. Examples and explicit rules move it a lot.

An output style is a saved instruction for tone and format that applies every time, so you stop re-typing 'short sentences, no fluff, no em dashes' in every prompt. Set it once and every answer starts from your defaults instead of the generic ones.

Write a short file that describes how you actually write - the words you use, the ones you ban, sentence length, how formal, how you open and close. Point Claude at it so your voice is a reference it reads, not something it has to guess each time.

text# my-voice.md
- Short, direct sentences. Lead with the point.
- Plain words over jargon. No buzzwords.
- Never use em dashes; use " - " instead.
- Warm, not stiff. Talk to one person, not an audience.
- Words I use: [list]. Words I never use: [list].

This is the strongest lever. Paste three short pieces you wrote and genuinely liked, and tell Claude to match their voice - not copy the content, match the rhythm, word choice, and structure. Examples carry a thousand implicit rules that no description captures.

1. 1Output style for the always-on defaults (tone, format, bans).
2. 2Voice file for the explicit rules and vocabulary.
3. 3Three examples for everything the rules cannot name.
4. 4Then review the first few outputs and add any rule you find yourself re-fixing.