Most people use AI to write LinkedIn posts by typing a topic and hitting generate. The output sounds generic, gets ignored, and they wonder why AI doesn't work for LinkedIn. The problem isn't the AI. It's the input.

The better approach is to reverse engineer what already works first, then train the AI on that structure. Here's the complete system.

Step 1: Find posts worth reverse engineering

You need posts that actually performed. Not posts that look good. Posts with real engagement data.

Where to find them:

Search your niche keywords on LinkedIn, filter by "Top" posts
Look at 5-10 creators in your exact space and sort their content by most commented
Track posts that keep getting reshared weeks after publishing (long-tail performers)
Screenshot and save any post that made you stop scrolling and actually read to the end

You're looking for 15-20 high-performing posts before moving to Step 2. Don't rush this. The quality of your source material determines the quality of your AI output.

Step 2: Extract the structural template using AI (not the content)

This is the step most people skip entirely. They copy the post, ask AI to "write something similar," and get a thinly disguised copy. That's not reverse engineering. That's plagiarism with extra steps.

The right approach: feed each high-performing post into Claude or GPT with this exact prompt structure.

What you get back is not a copy of the post. You get the blueprint behind it. Patterns emerge like:

Hook type: contradiction between common belief and personal result
Line 2: stakes establishment (what's lost if you ignore this)
Body structure: 3 specific examples with one-line explanations each
Proof placement: after the third example, not at the start
CTA: question that makes the reader self-identify

Do this for all 15-20 posts. After 10 analyses, patterns start repeating. Those repeated patterns are what actually drives engagement in your niche.

Step 3: Build your master prompt from the extracted patterns

Now you have 15-20 structural blueprints. Look for the 4-5 patterns that appear in at least 60% of your high-performing sample.

Combine them into a master prompt that includes:

The hook formula that kept showing up (e.g., "I [did X unconventional thing]. Here's what happened.")
The body pacing rule (e.g., one idea per line, no paragraphs over 2 sentences)
The proof format (specific number + specific outcome, not vague claims)
The CTA style (question that creates self-identification, not "follow me for more")
Character count range from your sample (probably 800-1,200)

This master prompt is now your trained content engine. It doesn't just know what good LinkedIn content looks like in general. It knows what good LinkedIn content looks like for your specific niche and audience.

Step 4: Add your voice layer on top

Here's where most AI content fails even with a good structural prompt. The structure is right. The voice is nobody's. It sounds like the average of the internet.

Fix this with style-priming. Paste 3-5 of your own best posts (even if they're old or underperformed) into the prompt before generating. Add this line:

The AI now has two constraints: the proven structure from reverse engineering AND your personal voice pattern. The output is structurally optimised and authentically yours.

Human-edited AI posts outperform raw AI posts by roughly 47% in engagement. Spend your editing time on the first 2 lines and the last line. Everything else is usually fine.

Step 5: Test, measure, and update the template

Run 8-10 posts using your master prompt over 30 days. Track which ones outperform.

When a post significantly overperforms, run it back through Step 2. Extract its structure. Check if it introduces a new pattern your template doesn't have yet. Update the master prompt.

Your prompt gets smarter every month because it's learning from your own real data, not from generic LinkedIn advice.

What this system produces after 60 days:

A content engine that generates first drafts in under 5 minutes
Posts that sound like you, not like a robot trained on motivational quotes
A growing library of structural templates specific to your niche
Engagement rates that compound because each post teaches you something new about your audience

The accounts generating 650k+ impressions in 6 weeks using AI aren't typing "write a LinkedIn post about leadership." They're running systems like this one.

What niche are you building content for? Curious whether the structural patterns I've seen hold across different industries or if they shift significantly.