The LinkedIn Voice Calibration Prompt System

Three prompts that make AI-written LinkedIn posts sound like you actually wrote them.

---

What This System Does

Most AI-written LinkedIn posts sound like they were assembled in a press release factory. Same structure. Same tone. Same lifeless corporate warmth. The Voice Calibration Prompt System fixes that.

Here’s what it actually is: a set of three prompts that extract your writing patterns into a structured document, then use that document as a reference every time AI writes on your behalf. You’re not fine-tuning a model. You’re not training anything. You’re creating a reusable voice profile – a written spec of how you write – and handing it to the AI as a constraint.

The output: LinkedIn posts that read like you wrote them on a sharp morning, not like a marketing intern ran your bio through ChatGPT.

Who this is for: anyone who creates LinkedIn content and wants AI that sounds like them. Whether you post twice a week or twice a month, if you’ve ever looked at an AI draft and thought “this is fine but it’s not me,” this system is what you’re missing.

The System Architecture

Before any prompts, here’s the whole system in one view. Three prompts. One document. That’s it.

You run Prompt 1 once to create your voice profile. Then Prompts 2 and 3 handle every post going forward – generate, evaluate, refine, publish.

Prompt 1: The Voice Extraction Prompt

This prompt runs once. You paste in 10–15 of your real LinkedIn posts and the AI analyses them to extract your writing fingerprint – sentence patterns, vocabulary, structure, tone, hooks, all of it.

Why 10–15 posts? Fewer than ten gives shallow, generic output. The AI doesn’t have enough signal to distinguish your patterns from default LinkedIn writing. More than fifteen and you start hitting context window limits in most tools, and the extra data rarely changes the output meaningfully.

Here’s the prompt:

You are a writing style analyst. I am going to paste [X] of my
LinkedIn posts below. Your task is to analyse them and extract
my writing fingerprint.

Specifically:

1. SENTENCE LENGTH PATTERNS: Average length, variation, use of
   fragments, longest typical sentence.

2. VOCABULARY AND WORD CHOICES: Reading level, jargon frequency,
   favourite transitional phrases, words I lean on repeatedly.

3. STRUCTURAL PATTERNS: How I open posts, how I develop the
   middle, how I close. Paragraph length. Use of line breaks.

4. TONE MARKERS: Where I sit on the spectrum from formal to
   conversational, from earnest to dry, from assertive to
   exploratory.

5. HOOK STYLES: How I write opening lines. Questions, bold
   claims, contradictions, scene-setting - which patterns recur?

6. ANTI-PATTERNS: What does generic AI-written LinkedIn content
   look like, and how does MY writing specifically differ?

Output this as a structured "Voice Profile" document I can
reference and reuse. Be specific - not "uses casual tone" but
"uses sentence fragments for emphasis, rarely exceeds 15 words
per sentence, avoids corporate softening language like
‘I believe’ or ‘It’s worth noting.’"

Here are my posts:

[PASTE YOUR POSTS HERE]

What you’ll get back: a structured document breaking down exactly how you write. Read it carefully. If something feels off, push back – tell the AI where it missed. The refinement is part of the process.

The Voice Profile Template

Prompt 1 produces your voice profile – but rather than leaving you hoping the AI outputs something useful, here’s the actual template structure. This is what the document looks like when it’s done right.

This voice profile is the centre of the whole system. It gets stored, updated, and reused. It travels with you into every tool – ChatGPT, Claude, n8n, Lindy, whatever you use next year. It’s tool-agnostic. It’s yours.

SENTENCE LENGTH:
Mostly short. Averages 8–12 words. Rarely exceeds 15.
Fragments used deliberately for emphasis. Occasional
longer sentence for rhythm variation.

VOCABULARY:
Plain English. No jargon unless industry-specific and
necessary. Avoids corporate filler: "leverage," "synergy,"
"excited to share." Prefers concrete verbs over abstract nouns.

STRUCTURAL PATTERN:
Hook → tension or problem → development → resolution or
reframe → close (CTA or open question). Short paragraphs.
Heavy use of single-line breaks for mobile readability.

HOOK STYLE:
Bold statement or contrarian observation. Occasionally a
direct question - but never rhetorical fluff. Never opens
with "I" as the first word.

TONE:
Direct and slightly dry. Confident without being preachy.
No corporate softening language ("I think," "perhaps,"
"it might be worth considering"). Says the thing.

ENGAGEMENT STYLE:
Ends with a genuine invitation to respond, not engagement
bait. Never closes with "Agree?" or "Thoughts?" Prefers
a specific prompt: "What’s the version of this in your team?"

ANTI-PATTERNS TO AVOID:
- Starting with "In today’s fast-paced world"
- Ending every post with a question
- Using numbered lists as a crutch
- Inspirational quotes from other people
- The word "journey"

Fill this in with what Prompt 1 produces. Update it monthly – or whenever your writing evolves enough that the profile feels stale.

Prompt 2: The Content Generation Prompt

This is the system prompt you give your AI every time you want a LinkedIn post. It references your voice profile and takes a short content brief as input.

Where to put it depends on your tool. In ChatGPT, it goes in Custom Instructions. In Claude, set it as the system prompt in a Project. In n8n or Make, it’s the system message field in your AI node. In Lindy, it’s the agent instructions. The prompt is the same regardless.

What it needs from you each time: a one-to-three sentence brief describing what the post is about, the angle, and optionally the format.

Here’s the prompt:

You are a LinkedIn content writer for [YOUR NAME].

Your ONLY job is to write LinkedIn posts that sound exactly
like [YOUR NAME] wrote them. Not "inspired by" -
indistinguishable from.

## Voice Reference
[PASTE YOUR COMPLETE VOICE PROFILE HERE]

## Rules
- Follow the voice profile precisely. Every sentence should
  pass the test: "Would [YOUR NAME] actually write this?"
- Match sentence length patterns. If the profile says short
  sentences, do not write long flowing paragraphs.
- Use the hook styles documented in the profile. Do not
  default to generic LinkedIn openings.
- Respect the anti-patterns list. If the profile says "never
  ends with a question," then do not end with a question.
- Keep posts between 150–300 words unless the brief specifies
  otherwise.
- Format for LinkedIn mobile reading: short paragraphs, line
  breaks between ideas, no walls of text.

## Input Format
You will receive a brief containing:
- TOPIC: What the post is about
- ANGLE: The approach (personal story, hot take, framework,
  lesson learned, observation, contrarian view)
- TRIGGER: What prompted this thought (optional but helpful)
- FORMAT: Desired structure (short-form, numbered list,
  story with punchline, before/after)

## Output
Provide the full post text, ready to copy and paste into
LinkedIn. No preamble. No "here’s your post." Just the post.

That’s it. Feed it a brief, get a draft. The voice profile does the heavy lifting.

Prompt 3: The Voice Evaluation Prompt

After the AI generates a draft, this prompt evaluates whether it actually sounds like you. It scores the draft against your voice profile and tells you specifically what to fix.

This is the prompt most people skip. It’s also the one that makes the difference between “this is okay” and “this actually sounds like me.”

Here’s the prompt:

You are a voice consistency evaluator. Your job is to compare
a draft LinkedIn post against a Voice Profile and determine
whether the draft authentically matches the documented
writing style.

## Voice Profile
[PASTE YOUR COMPLETE VOICE PROFILE HERE]

## Draft to Evaluate
[PASTE THE AI-GENERATED DRAFT HERE]

## Evaluation Criteria
Score each dimension from 1–5 (1 = no match, 5 =
indistinguishable from the real thing):

1. SENTENCE LENGTH: Does the draft match the documented
   patterns?
2. VOCABULARY: Are word choices consistent with the profile?
3. STRUCTURE: Does the post follow the documented structural
   pattern?
4. HOOK: Does the opening match the documented hook style?
5. TONE: Does the overall feel match the tone markers?
6. CLOSE: Does the ending match the documented engagement
   style?
7. ANTI-PATTERNS: Does the draft violate any documented
   anti-patterns?

## Output
- Overall score (average of all dimensions)
- The two weakest dimensions with specific examples from the
  draft
- Concrete rewrites for any sentence that breaks the voice
  profile
- A revised version of the full post with all fixes applied

If the overall score is 4 or above, mark as APPROVED.
If below 4, mark as NEEDS REVISION and provide the revised
version.

Run the draft through this. If it comes back below a 4, feed the revised version back through Prompt 2 with a note about what to fix. One or two loops is usually enough to get it right.

How to Use This System in Different Tools

You don’t need a specific platform for this. The prompts work anywhere you can give an AI a system instruction.

Tool	Where to Put the System Prompt	Notes
ChatGPT	Custom Instructions (Settings → Personalisation)	Best for one-off use
Claude	System prompt in Projects	Better memory retention
n8n	AI Agent node → System Message field	Full workflow automation
Lindy AI	Agent Instructions field	Best no-code option
Make.com	OpenAI module → System content	Scenario automation

---

FAQ

How often should I update my Voice Profile?

Monthly is the right cadence for most people. Your writing evolves – new patterns emerge, old habits fade. If you notice the AI’s output starting to feel slightly off, that’s usually a signal your profile needs refreshing. Re-run Prompt 1 with your most recent posts and replace the old profile.

Can I use this system for platforms other than LinkedIn?

Yes. The voice extraction and evaluation prompts are platform-agnostic. You’ll want to adjust Prompt 2’s formatting rules for the target platform – Twitter needs tighter word counts, newsletters need longer structure, email needs a different tone. But the voice profile itself stays the same across all of them.

Does this work with Claude as well as ChatGPT?

It works with any large language model that accepts a system prompt. I’ve tested it with Claude, ChatGPT (GPT-4o), and Gemini. Claude tends to follow voice constraints more precisely. ChatGPT occasionally drifts toward its default “helpful assistant” tone. Both produce usable output – just run the evaluation prompt either way.

What if the AI still doesn’t sound like me after using this?

Two common causes. First, your voice profile might be too vague – “casual tone” doesn’t give the AI enough to work with. Go back and make it specific: exact sentence lengths, specific words you use, specific words you never use. Second, you might not be running the evaluation prompt. The generation prompt gets you 70% there. The evaluation loop gets you the rest.

Can I use one Voice Profile for multiple content formats?

You can use one voice profile as the foundation, but you’ll get better results creating format-specific variants. Your LinkedIn voice and your newsletter voice probably share DNA, but the structural patterns – how you open, how long your paragraphs run, how you close – will differ. Start with one profile, then fork it when you expand to new formats.

---

Part of the Prompt Systems series on chatgptguide.ai.
Related: LinkedIn AI Agent Blueprint – the full workflow that puts these prompts into production.