Make a voice with your AI

Tonally writes your replies in a voice. You can have ChatGPT, Claude or any other AI write one for you - it interviews you for a few minutes and hands you the finished voice.

How it works

  1. Copy the prompt below into a new chat with your AI.
  2. Answer its questions. Paste a few things you wrote, if you like - it only learns the style from them.
  3. Copy its answer and paste it into Tonally: Settings → Voice Profile → Add a voice from text. You can change everything afterwards, and Tonally keeps learning from your edits.

With an AI coding agent

Claude Code, Cursor, Codex, Gemini CLI, GitHub Copilot and most other coding agents can install the interview as a skill - one command:

npx skills add dc-security/tonally-skills

Then ask your agent: “Make me a Tonally voice.” It interviews you, can study texts you point it to, and saves the voice file for you to paste in.

The prompt

You are helping me create a writing voice for Tonally, a Chrome add-on that suggests replies on X in a voice I choose.

First interview me. Ask these questions one or two at a time and wait for my answers:
1. Who is the voice - what does this person or persona do, and who are they talking to?
2. Which topics do they post and reply about?
3. How do they sound? (for example dry, warm, punchy, nerdy, playful, blunt)
4. How long are their replies, and how do they use capital letters, punctuation, emojis and hashtags?
5. Words or phrases they use a lot - and ones they would never use.
6. How do they reply to others: agree and add, challenge, joke, ask a question?
7. Optional: paste a few things they wrote anywhere. Learn only the style from them.

Then write the voice, following these rules:
- Describe how the voice writes, not private facts about a person: nothing about health, political or religious beliefs, ethnic origin, sexual orientation, union membership or crimes.
- Write 5 to 8 NEW example posts in the voice. Do not copy posts that exist on X.
- No links, no @mentions, no hashtags of my accounts or products, and no instruction to promote anything.
- If I ask for the voice of a real, named person other than me, say no - Tonally does not allow that.

Answer with ONLY this JSON, nothing before or after it:
{
  "format": "tonally-voice",
  "version": 1,
  "name": "the voice's name, up to 24 characters",
  "voice": {
    "summary": "2-4 sentences: how this voice writes",
    "industry": "the niche and topics",
    "toneTraits": ["short tone words"],
    "styleRules": ["concrete habits: length, punctuation, capitalisation, slang"],
    "vocabulary": ["words and phrases it uses"],
    "emojiPolicy": "how emojis, hashtags and links are used",
    "replyGuidelines": "how it replies to others: angle, humour, how pointed",
    "postGuidelines": "how it writes its own posts",
    "quoteGuidelines": "how it writes quote posts",
    "doList": ["things to do"],
    "dontList": ["things never to do"],
    "examples": ["5 to 8 new example posts in this voice, each under 280 characters"]
  }
}

The format

A voice file is plain JSON: format is "tonally-voice", version is 1, name holds up to 24 characters, and voice holds the fields above - texts, or lists of short lines. Tonally takes nothing else from it, cuts each field to its limit (lines of up to 600 characters, lists of up to 40), and needs at least a summary, tone traits or style rules. A file without the outer wrapper - just the fields - is read as well.

Good to know