What Actually Happens When I Build a Website with ChatGPT

I’ve found ChatGPT performs best when you tell it exactly what you expect in return. Clear instructions almost always produce better results than vague prompts.

The brochure wasn’t something to copy. It became evidence.

From that evidence, together with ChatGPT’s summaries and analysis, the website navigation emerged. Once the navigation made sense, every page had a clear purpose and the content naturally followed.

Before ChatGPT, this stage would have taken me hours.

Instead, I had a solid first draft of the information architecture within minutes.

That doesn’t mean I simply accepted every suggestion.

I challenged them, questioned them and refined them until they accurately reflected the organisation. Another thing I’ve found is that ChatGPT performs best when it has something concrete to work from. Research. Documents. Evidence. The more context I gave it, the stronger its output became.

Writing for the Design

Once the website structure had been agreed, ChatGPT’s role changed completely. This is probably the part people misunderstand most.

I wasn’t asking ChatGPT to write a website. I was designing the website first.

As I built each page in Elementor, I’d take a screenshot and paste it into ChatGPT with one instruction:

Fit the content into this design.

That meant keeping the logic, flow, SEO and word count while making the content work within the space I’d designed.

The conversations became very practical. My instructions sounded more like this:“This hero section only has room for twelve words, shorten that sentence and give back to me”, “These three columns need to balance visually, make text in each column exactly the same length but keep the logic”. Or “Make these two sections one”, and “This explanation needs another paragraph. Review the content already on the page for context, expand this section naturally, and avoid repeating ideas or information that have already been covered”.

The content wasn’t being written as isolated pages. It was being written for the design while supporting the overall structure of the website. So changing one section often meant refining another because every page needed to flow naturally into the next.

Clarity always came first. Once the message was clear, the SEO followed naturally.

Again, before ChatGPT, this stage would often have taken me days and could be incredibly frustrating. Instead, it became a rapid back-and-forth conversation that refined the website’s content in just a few hours. Perhaps surprisingly, it was also one of the most enjoyable parts of the entire project. 

ChatGPT didn’t perform one role throughout the project.

It kept changing jobs. First it was a researcher. Then an editor. Then a copywriter. Then a technical assistant. Finally, a proofreader.

Every time I reached a different stage of the project, it adapted to what I needed next.

It organised information, challenged my thinking, highlighted where a design needed more explanation and adapted content to fit the available space.

Every design decision was still mine. I simply reached those decisions much faster.

The content wasn’t being written as isolated pages. It was being written for the design while supporting the overall structure of the website. So changing one section often meant refining another because every page needed to flow naturally into the next.

Clarity always came first. Once the message was clear, the SEO followed naturally.

Again, before ChatGPT, this stage would often have taken me days and could be incredibly frustrating. Instead, it became a rapid back-and-forth conversation that refined the website’s content in just a few hours. Perhaps surprisingly, it was also one of the most enjoyable parts of the entire project. 

ChatGPT didn’t perform one role throughout the project.

It kept changing jobs. First it was a researcher. Then an editor. Then a copywriter. Then a technical assistant. Finally, a proofreader.

Every time I reached a different stage of the project, it adapted to what I needed next.

It organised information, challenged my thinking, highlighted where a design needed more explanation and adapted content to fit the available space.

Every design decision was still mine. I simply reached those decisions much faster.

Creating the Images

The illustrations supplied for the brochure were beautiful, but they weren’t designed for modern website layouts. Many were portrait illustrations that simply didn’t fit wide-screen designs.

This is where another AI tool entered the workflow.

Using Photoshop’s Generative Fill, I expanded each illustration beyond its original edges. Photoshop generated three possible versions, I chose the strongest one and prepared it for the website.

The result looked as though the artwork had always been created for the web. Generative Fill also proved invaluable later in the project when I separated elements from the original illustrations into individual layers. Those layers became the foundation of the subtle animated scenes on the homepage, allowing foreground and background elements to move independently while preserving the original artistic style.

Generative Fill wasn’t just used to extend the images—it also helped transform the original artwork into layered assets that could be animated on the homepage. 

Once the images were complete, ChatGPT generated descriptive filenames, alt text and image descriptions while I optimised and exported every image as lightweight WebP files. Like every other stage of the project, it was another back-and-forth collaboration.

Removing Friction During Development

ChatGPT continued helping long after the content had been written because development introduces a completely different type of problem.

When building WordPress websites I often begin with a professional theme before heavily customising it with CSS, animation and bespoke layouts. Every theme organises its settings differently. This is a great way to save clients a significant amount of development time and budget, starting with something that already fits the purpose before customising it.

But it’s a bit like driving a different car. You know how to drive it, but every manufacturer seems determined to hide the gas cap release somewhere different.

As I built the site, I’d constantly run into small technical questions.

“Where does the Litho theme hide this setting?” or

“What Litho theme master settings are overriding my CSS here?”

Instead of stopping to search documentation or forum posts, I’d ask ChatGPT and get an immediate answer, complete with step-by-step instructions tailored to the exact theme I was working on.

It also became an excellent CSS assistant, helping write, refine and troubleshoot hundreds of lines of code throughout the project.

So far more than a writing assistant.

By this point, ChatGPT had become my fastest technical reference, removing dozens of small interruptions that would otherwise have slowed development down.

Editing Before Launch

When the website was nearly finished, we reviewed it from two different perspectives.

I reviewed the design, checking spacing, typography, imagery, buttons, links, responsive layouts and the overall user experience.

ChatGPT reviewed the content, checking grammar, sentence flow, repetition, consistency, page hierarchy and whether the story made sense from beginning to end.

Those reviews weren’t competing with each other. They complemented each other. Together they produced a better website than either of us would have achieved alone.

Final Thoughts

One of the biggest misconceptions about ChatGPT is that it replaces the designer.

That has never been my experience.

Throughout this project it became a tool for speed, logic and problem-solving. It helped me understand complex information faster, organize it into a clear website structure, adapt content to fit my designs, prepare images for the web, remove technical roadblocks during development and polish the finished website before launch.

The creative direction, design decisions and final judgement remained mine from beginning to end. I never asked ChatGPT to design the website. Instead, it became another specialist sitting beside me throughout the project. The value isn’t in asking ChatGPT to do your job. It’s in knowing your job well enough to direct ChatGPT through every stage of it.

I don’t think that’s replacing web designers.

I think it’s changing—and dramatically enhancing—the way experienced designers work.

Looking Ahead 

One thing worth remembering is that this workflow will continue to evolve. The AI tools, applications and techniques described here will almost certainly look different in a year’s time, just as they’re different from those available only a year ago.

What won’t change is the process. The real value doesn’t come from any single AI tool—it comes from the conversation between human expertise and artificial intelligence. Asking better questions, refining ideas, testing alternatives and building on each result is what turns AI from a novelty into a genuinely useful creative partner. As the technology improves, that collaboration is only likely to become more capable.

One Last Question

After finishing this article, I asked ChatGPT one final question:

“What do you think of this article and the process it describes?”

Its answer was surprisingly close to my own.

“The article doesn’t argue that AI replaces expertise. It argues that expertise becomes more valuable because someone still has to guide the process, judge the output and make the final decisions. AI accelerated almost every stage of the project, but it never decided what the website should become. The conversation between designer and AI is what produced the final result.”

Of course, ChatGPT is designed to be helpful and generally cooperative, so I wouldn’t expect it to harshly criticise the process. My perspective comes from actually building the website. What I found interesting wasn’t that it agreed with me, but how closely its explanation matched what I’d experienced throughout the project.