A Client Story · Property Portal · 2024
450 Articles.
48 Hours.
Here's How.
The full story behind the AI content engine I built for Zimbabwe's largest property portal — and why it's not just "prompting ChatGPT."
The Client
Zimbabwe's largest property portal. Great traffic. Broken content strategy.
Propertybook.co.zw consistently ranked 2nd–6th on Google for 4,000+ real estate keywords. They had traffic, data, and domain authority. What they didn't have: any system for turning that advantage into content that ranked.
The Problem
One freelancer. One article. Two weeks. $200 each.
That was the old workflow. For a portal sitting on 4,000 high-intent keywords with no quality content ranking for them — it was like owning a gold mine and using a teaspoon to dig. The math didn't work. The content didn't scale. Google didn't care about generic AI text. They needed something entirely different.
The Brief
Not just fast. Actually good.
They were specific: no AI slop. Google's EEAT requirements (Experience, Expertise, Authority, Trust) meant generic generated content would actively hurt rankings. They wanted proprietary insights their competitors couldn't replicate — real property data, real locations, real market context. And they wanted a content management system to track all of it.
The Architecture — 6 Steps, Fully Automated
Keyword Analysis
GSC + ahrefs APIs identify the highest-ROI keyword opportunity from 4,000+ tracked terms.
Competitor Scraping
SerpAPI + Scrapestack pull the top-ranking articles and analyze content gaps.
Private DB Query
AI determines which proprietary metrics to pull from the client's 7,000+ property database via custom Coda API integration.
EEAT Assembly
Google Maps API + location research + dynamic writing guidelines selected by content type and topic.
Generation + QA
Claude generates the article in structured JSON. Automated QA checks for keywords, EEAT signals, brand voice, and accuracy.
CMS + Tracking
Status tracked from idea to publication. GSC dashboard monitors each article's traffic, rank, and CTR over time.
The Approach
Why I built this in low-code first.
Most developers would've started with React and Node.js. I started in Coda.io — building the entire 6-step AI pipeline as a low-code prototype in days, not months. That meant the client could see it working, test it with real data, and iterate before I committed to production infrastructure. By the time we went live, every edge case was solved. The system was battle-tested before it was even "built."
Articles. 48 Hours.
The system ran over a weekend. When the client opened their CMS on Monday morning, 450 fully written, EEAT-optimized, QA-checked articles were ready to publish — each with unique insights from their private property database that no competitor could replicate.
Average Google ranking increase.
Measured across all published articles over 6 months. The articles with proprietary database content outperformed everything. EEAT worked — but only because the system was designed to incorporate real signals, not fake them.
Genuinely amazing. Jheesh, well done. And thank you very much. Super chuffed with this.
— Client, Propertybook.co.zw
The Lesson
The best AI content isn't faster ChatGPT.
It's a system that knows things your competitors don't. Proprietary data. Custom quality controls. Dynamic writing guidelines. A feedback loop that measures what actually ranks. Any agency can generate AI text. What I built was a competitive moat — content that, by design, can't be replicated by someone who doesn't have access to that property database.