From ChatGPT to Codex: How I Started Automating My Website Updates

First Steps

When I first started using ChatGPT for website development, I was amazed at how easy it was. I could ask for HTML, JavaScript, SQL, CSS, or pretty much any code I needed, and within seconds it would generate it for me. I would then copy and paste the code into Dreamweaver, test it, and upload it to my website. Compared to writing everything from scratch, it felt almost magical.

For months I happily carried on working this way, convinced that this was as fast as website development could get.

Then, around three months ago, a message appeared suggesting that I try Codex.

I thought, "Why not?"

Enter the Codex

After a bit of experimenting, I created a folder on my computer, pointed Codex at my website files, and suddenly everything changed.

This was one of those genuine jaw-dropping moments.

Instead of generating code that I had to manually copy and paste, Codex could update the files directly. Better still, it could update multiple files at the same time. If I wanted to make a change across dozens of pages, it could work through them systematically and tell me exactly what it had changed. I could also control permissions to make sure important files were not modified accidentally.

Very quickly I realised this was far more than just a code generator.

As I spent more time reading social media posts about Skills, Markdown files, prompts, and AI workflows, I started wondering whether I could take things a step further.

The Magic of Visual Studio... With Codex

Could I create a system that automatically gathered information from Google Search Console, built a prompt from that data, and then used Codex to improve my website?

I was particularly interested in SEO.

With plenty of help from GPT, I started building Python scripts. I learned how to work with API tokens so that I did not need to keep logging into Google Search Console manually. The scripts could pull search performance data and save it into files that Codex could read.

From there, I created a dynamic prompt that used the latest search data to identify opportunities for improvement across my website.

To make the process even easier, I created a batch file that runs automatically once a month. The result is a workflow that gathers the latest search data, prepares the information, and gets everything ready for Codex to analyse and update.

What started as simply asking ChatGPT to write bits of code has evolved into a system that helps maintain and improve my website using real search performance data.

The exciting thing is that I feel as though I am only getting started.

The next stage is to turn everything I have learned into a training course so that other website owners, marketers, and developers can build similar AI-powered workflows for themselves.