Your LinkedIn posts sound like ChatGPT because you skipped the one step that matters. A voice prompt is a 500 to 800 word instruction that makes AI write like you, not like a corporate intern. Here's the exact template.
An AI voice prompt for LinkedIn is a structured instruction (500 to 800 words) you give ChatGPT or Claude before any writing task. It captures your sentence patterns, vocabulary, banned words, and worldview so the output sounds like you wrote it. Without one, every AI-written LinkedIn post sounds the same. Building one takes 90 minutes and requires 10+ existing writing samples. The DFY Voice Build does this for you in 3 days.
An AI voice prompt tells ChatGPT or Claude how you write, not what to write. It contains mechanical rules (sentence length, contractions, paragraph structure), banned words, signature phrases, and tone shifts by format. Build time: 90 minutes. You need 10+ writing samples. The result: LinkedIn posts that sound like you on first draft instead of requiring 30 minutes of editing per post.
A voice prompt is not a persona description. "Write like a friendly marketing expert" produces the same generic output as no instruction at all. AI cannot interpret vibes. It can follow rules.
A voice prompt is a set of mechanical writing rules the AI can actually execute. Average sentence length. Whether you use contractions. How long your paragraphs run. Which words you never use. Which phrases you repeat. What your worldview sounds like when compressed into 3 sentences.
Think of it this way. A recipe that says "make it taste good" produces nothing useful. A recipe that says "2 teaspoons salt, 1 tablespoon olive oil, medium heat for 4 minutes" produces a consistent result. Voice prompts work the same way. Specificity is the mechanism.
For the full breakdown of what a voice prompt is across all platforms, see what is a voice prompt. This post focuses specifically on LinkedIn because the platform has format constraints that change how the prompt needs to work.
Three reasons LinkedIn is harder for AI than other platforms.
The feed is saturated with AI content. LinkedIn's algorithm can detect AI-generated patterns. Posts that read like default ChatGPT output get up to 45% less engagement than human-written posts. A voice prompt breaks the AI patterns by forcing the output into your specific writing style instead of the ChatGPT house style.
Your audience reads you, not your brand. LinkedIn is personal. People follow a person's perspective, not a company's content calendar. If three posts in a row sound different from your usual voice, followers notice. They do not unfollow. They stop engaging. The algorithm reads the drop in engagement as a quality signal and reduces your reach.
The format is unforgiving. LinkedIn posts live or die in the first 2 lines (the hook before "see more"). Default AI writes weak hooks. A voice prompt with a hook-first rule and examples of your best-performing hooks produces dramatically better opening lines.
This is the exact structure we use in the DFY Voice Build. You can build it yourself using the instructions below.
Do not overthink this. The identity section sets the frame. Everything after it adds the mechanical precision.
This section does the most work. AI follows mechanical rules reliably. "Use 12-word sentences on average" produces a noticeably different output than "use 22-word sentences on average." Test both and pick the one that matches your actual writing.
This is where you kill the AI voice. Every AI has default filler words it reaches for. ChatGPT loves "navigate," "foster," "delve," and "it's important to note." Claude defaults to "I'd be happy to" and "certainly." Your banned list should include both AI defaults and industry jargon you personally avoid. For a starting list of marketing banned words, check the voice training guide.
These are the phrases, structures, or rhetorical moves you use repeatedly. "The contrast pair" (X is not Y, it's Z). "The specific number lead" (47 leads in 30 days). "The three-beat pattern" (You tried X. You tried Y. You tried Z). Look at your last 10 posts. You already have these patterns. Name them and put them in the prompt.
This section makes the generic voice prompt LinkedIn-specific. Without it, AI writes in paragraph form that looks terrible in the LinkedIn feed.
Your worldview is what separates your content from everyone else's. "I believe systems beat prompts." "I believe most marketing advice is designed to sell courses, not produce results." "I believe solopreneurs should own their content systems, not rent them." These beliefs colour every post the AI writes.
The examples are the final calibration layer. The AI uses them as ground truth for the rules above. Pick posts that got strong engagement AND sound like you at your best. Do not pick posts you wrote on a bad day.
The DFY Voice Build analyses your existing writing, extracts all 7 sections, and hands you a finished voice prompt plus custom GPT, hook library, and batching workflow. $497 at founder pricing. You own every asset.
See The Voice Build1. Vague descriptors instead of rules. "Write in a conversational tone" is not a rule. "Use contractions always. Average sentence length: 11 words. Never open with a question" are rules. AI cannot interpret "conversational." It can count to 11.
2. Prompt is too short. Under 300 words and the voice prompt does not contain enough signal to override the AI's default patterns. The sweet spot is 500 to 800 words. Over 1,000 and the AI starts ignoring instructions at the end of the prompt.
3. No banned words list. Without an explicit banned list, AI reaches for its defaults. Every ChatGPT post will include "it's important to note," "navigate the landscape," or "take it to the next level" unless you ban them. The banned list is the most underrated section of a voice prompt.
4. No real writing samples. Building a voice prompt from how you want to sound (aspirational) instead of how you actually sound (real) produces a mismatch. The AI matches the prompt, which does not match the person. Always build from real samples.
Run this 3-post test. Ask the AI to write 3 LinkedIn posts on different topics using your voice prompt. Print them. Hand them to someone who reads your content regularly. Ask: "Which of these did I write?" If they cannot tell, the prompt works. If they immediately flag one as "not you," the prompt needs another revision pass.
The other test is simpler. Read the output aloud. If you would not say those words in a conversation, the prompt is off. Voice prompts should produce text you could read on a podcast without sounding like a different person.
An AI voice prompt is a 500 to 800 word instruction you paste into ChatGPT or Claude before asking it to write LinkedIn content. It contains your mechanical writing patterns (sentence length, contractions, paragraph structure), your banned words, your signature phrases, your worldview, and your tone shifts by content format. Without it, every AI-written LinkedIn post sounds the same. With it, the output matches your voice at roughly 80% accuracy on first draft.
90 minutes if you have 10 or more existing LinkedIn posts or writing samples to analyse. The process is: collect samples (15 min), extract patterns (30 min), write the prompt (30 min), test and revise (15 min). If you have fewer than 10 samples, write 10 raw posts first. The voice prompt needs real writing patterns to capture. You cannot build a voice prompt from aspirational descriptions of how you want to sound.
Three common causes. First, the prompt uses vague descriptors instead of mechanical rules. "Friendly and conversational" tells the AI nothing actionable. "12-word average sentence length, always use contractions, never start with a question" gives it rules to follow. Second, the prompt is too short (under 300 words). Third, the prompt does not include banned words or phrases. AI defaults to corporate filler unless you explicitly ban it.
The core voice section (mechanical patterns, worldview, banned words) stays the same across platforms. But LinkedIn has specific format rules that other platforms do not: hook-first opening, line breaks after every 1 to 2 sentences, no hashtags in the body (3 max at the end), 1,300-character sweet spot for feed posts. Add a LinkedIn-specific format section to your voice prompt or create a separate LinkedIn wrapper prompt that references your core voice prompt.
The DFY Voice Build analyses your writing, extracts your patterns, and delivers a finished voice prompt, custom GPT, hook library, and batching workflow in 3 days. $497 at founder pricing. You own every asset forever.
See The Voice Build