We had two years of LinkedIn posts sitting in our own history and had never actually looked at them. So we pulled all 558, handed them to Claude, and asked it to tell us the truth about our own writing: what worked, what didn't, and why. Then we turned it into a skill that writes with us. Watch exactly how it went down below, then stick around, because the report changed something we thought we already knew about ourselves.
WATCH THE DEMO HERE
00:00 Introduction to Content Engines
01:07 Scraping LinkedIn Data with Apify
04:47 Analyzing Performance and Style
09:45 Building the AI Content Skill
14:25 Testing the Post Generator
QUICK SHOUTOUT
TestFit is the real estate feasibility platform for developers, architects, and contractors who need to know if a site pencils before they spend weeks drawing it. Founded by Clifton Harness, who started the company after growing frustrated counting parking stalls and drawing unit layouts by hand at his own development job. TestFit's generative design engine now evaluates 1,500+ deals a week for more than 6,600 users, and picked up the PropTech Breakthrough Award for Overall Construction Tech Solution in 2023.
What makes it different is that it treats site planning as a search problem, not a drafting problem. Feed it a parcel and your program requirements, and it generates and scores dozens of buildable layouts, complete with parking counts and yield on cost, in the time it used to take to draw one option by hand.
AI for CRE Collective members get a free pro forma add-on for both Site Solver and Parking Solver. Details located in the AI for CRE Tool Database.
WORKFLOW OF THE WEEK
We turned 558 of our own LinkedIn posts into a skill that writes like us
We used Apify to scrape two years of our own LinkedIn history, handed the export to Claude, and had it build a skill that studies real performance data before writing a single word for us again.
What you'll need:
An Apify account (LinkedIn post scraper actor, pay-per-use, our run cost about $1.02 against a $5 free credit)
Claude Code or Claude desktop with file access
Your own LinkedIn profile URL
At least a year of posting history to make the pattern real
Step 1: Scrape your own feed
We went to the Apify Store, searched "LinkedIn," and picked a LinkedIn profile post scraper. We dropped in one URL: our own profile. Set the post limit high, hit Start, let it run. Three minutes later we had every post exported as a CSV.
Why this matters: you are not buying data. You are buying your own receipts. Every hook that worked and every post that flopped is already sitting in that CSV, you just have never had a way to read it at scale.
Step 2: Make Claude study you before it writes anything
We dragged the CSV into Claude and gave it the instruction, close to verbatim:
"I have just attached an Apify LinkedIn Post Scraper Report... I want to know my writing style. I want to know what hooks perform the best... I want you to come up with a very comprehensive report... Ask me questions before getting started."
It asked four questions before touching the data: whether to include reposts (no, originals only), how to rank performance (both raw engagement and normalized for follower growth, since old posts with fewer followers were getting unfairly buried), where to show the output (inline in chat), and how deep to go on writing style (maximum). That last answer matters most. A shallow style breakdown gives you generic advice. A maximum-depth one gives you your own signature phrases back.
It came back having read 558 original posts going back to late 2024, broken down by format, hook type, comment-to-like ratio, and posting cadence, with the honest caveats included. The full analysis prompt and the exact question set are posted inside the Collective this week, so you can skip straight to asking your own Claude the same four questions.
Step 3: Turn the report into a skill, not a one-time PDF
A report you read once is advice. A skill you trigger every time is a system. We told Claude to build a LinkedIn post creator skill: save at least 20 of our best posts as reference templates, write a standard operating procedure grounded in the data from Step 2, and package it so it runs the same way every time we say "let's turn this into a LinkedIn post."
It asked more questions before building: what to name the skill, whether to wire it into any existing system (no, fully self-contained for the demo), how to choose the 20 reference posts (pure top 20 by engagement), and what the skill should hand back when triggered (two to three options, not one). The finished skill file, SOP and all, is posted inside the Collective too, ready to drop into your own Claude setup.
Step 4: Feed it something raw and watch it work
We restarted Claude so the new slash command would load, then triggered the skill with almost no input: a single news article about AI firms buying up Manhattan office space. No angle, no draft, nothing.
The skill loaded its two evidence files (the post analysis and the reference templates) on its own, then handed back three post options and a scorecard recommending which one to run with, before we typed a second sentence.
Time comparison: the whole build, scrape to a working skill, ran in 14 minutes and 41 seconds on camera. Manually reading and pattern-matching 558 of your own posts for hooks, cadence, and voice is not something anyone actually does, which is why most brand voice guides are a paragraph of adjectives instead of data. It's one of two new workflows we post inside the Collective every week.
THE EXECUTION ASSET LIVES INSIDE THE COMMUNITY
The full analysis prompt, the skill-build prompt, the question set, and the finished LinkedIn post creator skill file from this build are posted inside the AI for CRE Collective, ready to adapt to your own post history. We have hundreds of step-by-step workflows with demo videos, copy-and-paste prompts, and downloadable skills inside the Collective, and we post two new workflows every week.
Standard: $279/year or $79/month
Premium: $799/year or $199/month
WHAT THIS IS / WHAT THIS ISN'T
What this is: a way to ground your AI-written content in your own actual performance data instead of a generic prompt like "write in a professional but engaging tone." The skill gets more accurate the more history you feed it, and it gets better every time you add a new post to the reference set.
What this isn't: a publish button. Every option the skill hands back still needs a human read before it goes live. Our own first output needed real edits before it sounded like something we would actually post. It also only knows what happened in the past. It cannot tell you a hook will work, only that a hook like it has worked before. And if you have fewer than a year of posts, or you have mostly reposted other people's content, the pattern it finds will be thin.
BONUS
And one more thing. When we asked Claude to rank our posts by raw engagement versus normalized for follower count, it flagged something we did not expect: our best-performing post by raw likes was from three months ago, but our best-performing post by normalized engagement was from over a year ago, back when we had a fraction of the followers. The old post was actually the better piece of writing. Raw numbers had just been hiding it under an unfair sample size. That is the kind of thing you only catch when a model is willing to do the math you would never do by hand.
TRY IT YOURSELF
Export your own LinkedIn post history and hand it to Claude with one instruction: analyze it thoroughly before writing anything. Get started on Apify.
WHAT'S COMING UP
Using AI to Find & Screen Land for Development
Workshop
Wednesday, August 19
9:00 AM PT / 12:00 PM ET
$199 (free for Collective members)
Use Claude and public parcel data to find, screen, and pursue development sites across residential, multifamily, industrial, retail, and mixed use projects.
WHO WE ARE
We run the AI for CRE Collective: where 830+ CRE professionals learn to implement AI, with hundreds of video walkthroughs, copy-and-paste prompts, downloadable skills, 700+ AI tools in the database, and so much more.
If you'd like more hands-on help, we work with enterprise and institutional clients helping them learn, utilize, and implement AI into their business.