The First Version Was the Easy Part: How I Used ChatGPT to Build CraigMeyer.org

A working website arrived surprisingly quickly

I began by describing my goals and the kinds of content I wanted to share. ChatGPT used that direction to create the first working version of the site. The speed surprised me. What might once have required a long stretch of coding, searching for templates, and experimenting with design appeared in usable form very quickly.

That first result mattered because it changed the starting point. I was no longer staring at a blank page. I had something concrete to inspect, question, and improve.

But it was not a one-prompt success story. A working first version is not the same as a finished website, and a finished website is not the same as a dependable way to manage one.

From there, ChatGPT helped me revise the code and layout, organize and refine content, develop visual concepts, generate images, research factual claims, and manage versions through Sites. ChatGPT officially supports generating and editing images, while Sites can create, save, refine, and deploy hosted websites. Those capabilities gave me a much broader creative toolkit than my coding skills alone would have provided.

My responsibilities did not disappear. I still had to define the site’s purpose, decide what belonged on it, evaluate whether the design felt right, check facts, correct mistakes, and approve what other people would eventually see. ChatGPT could produce options at remarkable speed, but it could not decide what CraigMeyer.org should represent. That remained my job.

The harder problem was creating a reliable process

The most demanding part was not generating pages. It was establishing operational controls for continued work with ChatGPT.

My project-management experience taught me to look beyond the immediate deliverable. If I planned to keep adding articles, projects, images, and updates, I needed a process that would reduce confusion and prevent avoidable mistakes.

Questions emerged. Which conversation should be allowed to change the website? How should approved content move into the site? Which version was current? How could I review the exact result before it became public?

Those questions led me to create clearer roles, shared instructions, content handoffs, approval points, version checkpoints, and a rule that only one website update should be integrated at a time. I also separated approval of an article from approval to publish the website version containing it.

That distinction is especially important with Sites. OpenAI’s documentation explains that a saved version can serve as a reviewable candidate and that deployment is the separate step that makes the selected version available at a production URL. In practical terms: save first, inspect the actual result, and publish only after it passes review.

I did not invent these controls because ChatGPT had caused one dramatic failure. I created them because good project management anticipates predictable risks before they become emergencies. AI can accelerate production, but acceleration makes review and change control more important, not less.

What I learned about building with AI

The experience changed how I think about AI-assisted work.

First, AI lowers the barrier to creating something ambitious. I could combine my basic web knowledge, professional ideas, and judgment with capabilities I did not possess at a professional level. That did not make me a professional developer or designer. It let me direct and evaluate work that would otherwise have been much harder to produce.

Second, prompting is only one part of the process. Clear instructions matter, but so do source material, follow-up feedback, defined boundaries, and careful review. OpenAI’s own prompting guidance recommends explaining the desired result, audience, source material, and review process—and reviewing important work before sharing it.

Third, AI does not remove ownership. If a fact is wrong, a page is confusing, an image is inappropriate, or a change is published too soon, I cannot reasonably blame the tool. The website carries my name, so the final accountability is mine.

Finally, process does not have to suppress creativity. In this case, it gives me more freedom to experiment. I am more comfortable testing a new article, design, or image when I know the current site is protected and nothing will become public without a deliberate review.

My practical checklist

If I were beginning another website with ChatGPT, I would use these principles from the start:

1. Define the purpose and audience before asking for code. A fast build is useful only if it is building the right thing.
2. Treat the first working version as a prototype. Use it to make decisions, not as proof that the work is finished.
3. Review content, design, facts, and functionality separately. A page can look polished while still being inaccurate or difficult to use.
4. Give different conversations clear roles. Keep research, writing, visual development, and website publishing focused, while using shared project instructions where consistency matters.
5. Verify claims, links, and asset rights. AI-generated confidence is not evidence.
6. Control changes. Work on one update at a time, preserve a restorable version, and document what changed.
7. Review the exact website version before deployment. Approval of the words is not automatically approval of the rendered page.
8. Keep the human owner in charge. AI can propose and produce; the owner must judge and approve.

The real result

CraigMeyer.org is both a professional home for my work and an ongoing AI experiment. ChatGPT helped me create it more quickly—and with stronger visual design—than I could have managed alone. More importantly, I learned how to combine that creative speed with human judgment and project discipline.

The lesson is not that anyone can type one prompt and receive a perfect professional website. My experience suggests something more useful: a person does not need to possess every technical and creative skill personally to build something meaningful. But the better the AI becomes at producing work, the more important it is for the person directing it to provide purpose, standards, scrutiny, and accountability.

The first version may be the easy part. Building a trustworthy way to improve it is where the real work—and much of the learning—begins.