The reference document that teaches any AI tool to write in your voice. Definition, structure, examples, and how to use one across ChatGPT, Claude, and Gemini.
A voice prompt is a 500-800 word reference document that captures how a specific person writes — sentence patterns, vocabulary, banned words, tone shifts, signature moves — fed into an AI tool before any task to produce output that sounds like that person rather than generic AI default.
Out of the box, every AI tool defaults to the same voice: polished, professional, characterless. Public reporting from LinkedIn editorial and platforms like BrightEdge consistently shows that AI-assisted content matches generic templates more than 70% of the time when no voice context is provided. The result is content that ranks the same as everyone else's, sounds like everyone else's, and converts at default rates.
A voice prompt is the structural fix — the same systems-over-prompts principle applied to voice. It does for AI what a brand voice document does for human writers — but more specific, because the AI needs mechanical rules, not vibes.
Five sections, in order:
The Voice System Playbook walks through a complete build, from raw writing samples to finished voice prompt. 30 pages, every prompt we use, a real case study. Free.
Get the Free PlaybookChatGPT. Three options. Paste into Custom Instructions (Settings → Personalization). Use as the first message of a fresh conversation. Or build a Custom GPT with the voice prompt in the instructions field — most reliable approach because it can't be forgotten mid-conversation.
Claude. Use Claude Projects: paste the voice prompt as the project's system prompt. Every conversation in the project inherits it. For one-off tasks outside a project, paste it as the first message. See the ChatGPT vs Claude for LinkedIn comparison for which engine fits which format.
Gemini. Use the Gems feature (custom assistants) and paste the voice prompt into the instructions. Works the same as Custom GPTs.
Other tools. The principle is universal: voice prompt first, task second. If the tool accepts a system prompt, the voice prompt goes there. If it doesn't, paste it before every task.
These three terms get confused. They're different layers.
The DIY workflow (full walkthrough: how to train ChatGPT on your writing style): gather 10-20 pieces of your existing writing, paste them into ChatGPT or Claude with the prompt "Analyse these samples and identify mechanical voice patterns: sentence length, vocabulary level, paragraph structure, recurring phrases, and what makes this voice distinct from generic professional writing." Edit the AI's analysis into the five-section structure above. Test it. Iterate.
Full step-by-step instructions: how to make ChatGPT sound like you. Free playbook: The Voice System Playbook.
Pay (DFY) if: your hourly rate is over £100 and the 4-6 hour build is worth more in opportunity cost than the £497 service fee, or you've tried DIY and the output still doesn't sound like you. DFY Voice System uses The Voice Build methodology — same process documented in the playbook, executed for you in 2-3 working days. You own every asset.
DIY if: you have a focused weekend, want to understand the process before buying anything, or your writing samples are sparse and you'd rather develop the voice as you go.
The free Voice System Playbook has every prompt and the full five-section structure. 30 pages. One weekend. Your voice, in the machine.
Get the Free PlaybookDFY Voice System reverse-engineers your existing writing into a voice prompt + custom GPT + workflow. £497, delivered in 2-3 working days. You keep every asset forever.
See The Voice BuildA voice prompt is a 500-800 word reference document that describes how a specific person writes — sentence patterns, vocabulary, banned words, signature moves, tone shifts by format. It's pasted into an AI tool before any task so the output sounds like that person rather than generic AI default.
500 to 800 words. Shorter than 300 is too vague; longer than 1,000 eats into context window. The Voice Build methodology targets 500-700 words.
Five sections: voice essence (one-paragraph description), mechanical rules (sentence length, paragraph length, contractions, punctuation), banned words (phrases never used), tone by context (how voice shifts by format), and signature moves (3-5 distinctive habits).
Paste it into Custom Instructions, use it as the first message of every conversation, or build a Custom GPT with it in the instructions field. The Custom GPT route is most reliable because the voice can't be forgotten mid-conversation.
A brand voice document is for humans (style guide). A voice prompt is for AI (mechanical instruction set). The voice prompt is more specific and less abstract.
Yes. DIY takes 4-6 focused hours using existing writing samples. The Voice System Playbook documents the process for free. DFY costs £497-997 and saves the setup time.