How I built my own WhatsApp outreach tool for the secondhand market — with Claude and ChatGPT
When you start working with resales in Dubai — one of the first tasks is cold outreach to sellers. The database is huge, you can’t handle it manually, and ready-made WhatsApp broadcasting services are either expensive, or quickly block accounts, or don’t give you the control you need over the copy and sequence.
So I decided to build my own tool. Almost all the code is a joint effort with Claude Code and ChatGPT Codex. I set the task, they wrote, I refined the logic and tested it. That’s roughly what product development looks like when you’re a designer, not a developer — but you know exactly what you want to get.
How it works
Parsing and prioritizing the database
The Python script reads the secondary-market listings database and ranks who to message first. The logic takes into account property type, area, listing age, and other parameters. Not just a queue, but a smart selection.
Personalized copy
Each message is generated from a template — but not copy-pasted. The script varies the wording, inserts details about the specific property, and changes sentence structure. WhatsApp can detect identical messages and ban accounts — personalization isn’t about politeness, it’s about keeping the tool alive.
Sending waves
Messages are sent in waves of 40. There is a pause between waves. This mimics human behavior and reduces the risk of a ban.
Automated Chrome
The script launches a Chrome session with WhatsApp Web, inserts the text into the right chat, and opens it. All I have to do is hit Send. This matters — full automation of WhatsApp sending isn’t possible without the risk of an immediate block, so the final step is always left to a human.
CRM and mobile version
In parallel with the broadcasting tool, I built a CRM to track all the waves. I can see who I messaged, when, what the text was, and whether they replied.
There’s a web version and a mobile one — so I can keep sending from my phone when I’m away from my laptop.
Everything is hosted on Cloudflare Worker. Sign-in is via a magic link: enter your email, get a code by email, and you’re in. No passwords to remember.
What this means for me as a broker
The tool saves me several hours a day and removes the most annoying part of the job — monotonous manual work. Instead, I spend that time talking to the people who replied.
But what’s more important: I built it myself. I didn’t hire a developer or buy something off the shelf. I took Claude Code and Codex, described the task, iterated, tested, and added features as I figured out what was needed.
This is what I call design with the help of machine intelligence — not “press a button and get a product,” but a collaboration where you handle the logic and quality, and the tool writes the code.
The product is alive, and I’m continuing to develop it.