Open a Google Doc. Write 2 things.
1st, a do's and don'ts list. Short sentences or long? Contractions or not? What words would you NEVER say in a million years? Then write 1 bad example of your brand and one good one. AI learns faster from examples than rules.
β "A new feature was launched to improve operational efficiency."
β
"We shipped a small fix that saves teams hours every week."
2nd, 1 paragraph about your reader. Not a 10-page persona. Just who they are, how they think, and what they're allergic to. "Marketing leaders at growth-stage companies. They hate slop. Speak like a friend, not a coach." That's it.
Then start a banned word list in the same doc. Every time AI outputs something that makes you wince, add it. "Innovative." "Game-changer." "Synergy." "Robust." The list compounds and your output gets sharper on autopilot.
What to know on tooling: use Claude to actually BUILD the voice doc. It's better at capturing nuance, picking up on subtle tone differences, and writing in a way that sounds human. ChatGPT is great at executing once the voice is defined. So use Claude to create the doc, then load it into ChatGPT Custom Instructions or a Custom GPT so it runs automatically on every conversation.
Prompt to generate your voice doc in Claude:
"Let's define my brand voice. Begin by clarifying its purpose, identifying an appropriate tone, and understanding my natural communication style. Ask me insightful follow-up questions and challenge any responses that seem too generic. After defining my voice, draft a concise 1-page guide I can apply universally."
Then follow up with:
"Act as an expert brand strategist and copywriter. Use the brand voice we just defined. Ensure consistency in tone, rhythm, and language. Challenge any outputs that seem generic or off-brand."