Introduction

Even the best LLMs have no idea how you want to sound. If you don’t define your style, the output will be “correct but generic”. Below is the end-to-end process we run with clients at Macai Studio—finish it solo in 1–2 hours and you’ll own a “magic block” (a permanent system prompt). Paste it once, and every future generation will read as if you wrote it yourself.

Step 1. Gather raw stylistic material

• Collect 10–30 short or 3–5 long pieces of your own writing: business e-mails, posts, docs, chat logs, comments.

• The more contexts (pitch, complaint, thank-you, tech explanation), the better.

• Strip everything except the body text: no sender blocks, timestamps, third-party quotes.

• Save as clean .txt or .md files in one folder—makes later copying painless.

Step 2. Use that prompt

Use that prompt, pasting after that prompt your raw messages or file with them.

EXTRACT MY STYLE
TASK: Based exclusively on the provided texts, perform a thorough stylistic analysis and generate a comprehensive guideline. This guideline will serve as a permanent system prompt for any Large Language Model (LLM) to replicate my writing style.

OUTPUT REQUIREMENTS:

Document Title: "GUIDELINE – My Writing Style"

Format: A list of direct, actionable commands using "Always..." and "Never..."

Length: Approximately 250-400 words.

Language: English, but preserve all Polish names, phrases, and grammar as they appear in the source texts.

Structure: Cover all sections using the exact Markdown headings below. Analyze the provided examples to derive the rules for each section.

1. TONE & SENDER POSITION
Analyze the relationship between the writer and the recipient (e.g., formal, informal, authoritative, friendly).

Describe the overall emotional tone (e.g., confident, empathetic, direct, casual).

Identify the sender's role or persona (e.g., expert, collaborator, peer).

2. OPENING & CLOSING FORMULAS
Extract specific phrases used to start and end messages (e.g., "Hi [name]," "Best regards," "Cheers").

Identify any characteristic sign-offs or salutations.

3. LINGUISTIC CORRECTNESS
Detail the observed grammar, spelling, and punctuation habits.

Note any deliberate deviations from standard rules (e.g., use of sentence fragments, lowercase 'i').

Specify any unique use of capitalization or punctuation (e.g., using hyphens, em-dashes, or bolding).

4. STRUCTURE & FORMATTING
Describe how paragraphs are structured (e.g., short paragraphs, single sentences).

Note the use of lists, bullet points, or other formatting to improve readability.

Identify the typical length and flow of the messages.

5. VOCABULARY & FAVOURITE PHRASES
List specific words, idioms, or colloquialisms that appear frequently.

Identify any preferred jargon or technical terms.

Note the use of emojis or slang.

6. CONTEXT VARIANTS
Based on the provided examples, generate rules for how the style shifts depending on the message's purpose (e.g., when making an offer, sending a payment, or providing a technical explanation).

Highlight any patterns in how the tone or formality changes for different contexts.

7. CONTACT FOOTER
If a contact footer is present in the input texts, describe its typical contents and formatting. If none is provided, state this.

SUMMARY:

At the very end, add the following one-sentence summary: "This prompt ensures every generated message sounds exactly like [your name], with flawless Polish and a consistent personal brand voice."

PROHIBITIONS:

Do not use generic words like "user," "client," or "person" instead of my actual name.

Do not introduce any spelling or grammar rules that are not directly supported by the provided texts.

Do not add any phrases or vocabulary that I have not used.

Do not include any commentary or explanations outside of the generated guideline itself.

INPUT: Only the texts I provide below the line.

--- END OF INSTRUCTIONS ---

Step 3. Plug-and-play paths for every major LLM & automation platform

Below are copy-paste-ready places to drop your final SYSTEM-PROMPT so that every AI answer sounds exactly like you from day one.

Injecting your voice into n8n (automation + AI Agent)

Below is a copy-paste recipe that lets every e-mail, Slack message or SMS generated by an n8n workflow sound 100 % like you.

Use it with n8n

1. One-time prep

a) Store the final SYSTEM-PROMPT in n8n Credentials
Settings → Credentials → Add → type “Generic” → name it my_style_prompt
Paste the whole prompt into the single field → save.

b) (Optional) Keep the prompt under 400 words to fit inside memory-limited cloud nodes.

2. Classic workflow (no Agent) – “Generate & send”

Trigger (any) → OpenAI / Anthropic / Ollama node

• Resource: “Message”

• Operation: “Create a text completion / chat message”

• Model: gpt-4-turbo / claude-3-haiku / llama3 …

• System Message: press “Expression” icon
  ={{$credentials.my_style_prompt}}

• User Message: Write a short payment-reminder e-mail to {{$json.clientName}}
  ↓
  Gmail / Outlook / Slack node → send the returned text.

That is it—every run uses your voice without extra clicks.

3. AI Agent workflow (n8n AI Agent node ≥ 1.28)

a) Create a new workflow → add AI Agent node
b) Agent Provider:
– OpenAI, Anthropic, Ollama, or local LM-studio (any chat model)
c) System Prompt field → Expression
={{$credentials.my_style_prompt}}
d) Tools (optional):
– Read PDF invoice
– Query CRM for due date
– Send Gmail
e) Chat / Webhook trigger → ask “Remind client X about payment”
Agent returns text in your style and can call “Send Gmail” tool automatically.

ChatGPT (web, iOS, Android)

a) Custom Instructions
Settings → Personalization → Custom Instructions
Top box: paste the whole SYSTEM-PROMPT.
Bottom box (optional): short context, e.g. “Freelance web-dev, Polish & English business e-mails.”
Save → works globally, no further action needed.

b) Custom GPT
Explore GPTs → “Create” → “Instructions” tab → paste SYSTEM-PROMPT.
Add logo, name, knowledge files (price list, FAQ) → publish link to team or clients.

c) OpenAI API
messages=[ {"role": "system", "content": "<your prompt>

Claude (Anthropic)

a) Project Instructions
New Project → “Project Instructions” → paste SYSTEM-PROMPT → toggle ON “Enable for all chats”.
Works on web & mobile instantly.

b) API / SDK
system="<your prompt>

Google Gemini / Vertex AI

a) Gemini Advanced web → Settings → “Custom response style” → paste prompt.
b) Vertex AI REST → "system_instruction": "<your prompt>

Microsoft Copilot ecosystem

No global custom prompt yet → work-arounds:

• Copilot Studio → create new topic → “System message” → paste prompt → publish.

• Power Automate → “Create text with GPT” action → put SYSTEM-PROMPT inside “Instructions” field.

LangChain / LlamaIndex / crewAI (Python)

with open("moj_styl.txt") as f: SYSTEM_PROMPT = f.read()

chain = ( SystemMessage(content=SYSTEM_PROMPT) + HumanMessagePromptTemplate.from_template("{query}") ) | chat_model

crewAI: agent = Agent(role="Copywriter", backstory=SYSTEM_PROMPT, goal="Write exactly like the user")

Notion AI / Slack AI / ClickUp Brain

Notion:
Type “/AI” → choose “Custom” → paste SYSTEM-PROMPT → save.
Now every AI generate inside Notion uses your voice.

Slack:
Slack AI settings → “Custom instructions” → paste prompt → save.
Use /ai write … → output matches your style.

ClickUp Brain:
Space settings → “AI prompts” → “Add” → name it → paste SYSTEM-PROMPT.
Inside any task: @Brain write … using prompt “My voice”.

No-code automations (Make.com, Zapier)

Trigger: “New row in Google Sheet (client data)”
Action: OpenAI / Claude module → “System Prompt” field = SYSTEM-PROMPT
Action: Gmail / Outlook → send generated text.
Result: fully branded e-mail created & sent in <30 s.

Local models (Ollama, LM Studio, GPT4All)

Ollama one-liner: ollama run llama3 /set system "<paste prompt>

Permanent image: echo "FROM llama3\nSYSTEM """<prompt>

Security & version control

• Store prompt in encrypted repo (private GitHub) or company Key-Vault.

• Use env-var SYSTEM_PROMPT_VERSION – rollback in seconds.

• Quarterly: review 5 latest human-written texts → update prompt → tag new version (v1.3).

With these paths you get one voice, unlimited channels – paste the SYSTEM-PROMPT once, and every “Generate” button returns text indistinguishable from your own.