How I use AI to build real websites
Douglas Pan ·
I use AI in three moments: structuring content, generating the first version of the code and reviewing repetitive details. Everything goes through human review before going live. The model speeds up execution; the judgement about what belongs on the page stays mine.
What AI does well
Models are great at turning an already-made decision into working code. When the page architecture is clear, generating components, states and variations takes minutes.
They also help with repetitive tasks nobody enjoys: per-route metadata, structured data, image alt text, versions in another language.
What stays human
Deciding what the page needs to say. Choosing what to cut. Defining visual hierarchy and rhythm. Checking whether the copy sounds like the brand and not like a generic model.
Fact-checking is human too. A model confidently writes about results that never happened — that's why no number goes on the site without a source.
The workflow in practice
Diagnosis and content architecture first, always in plain text. Then AI-assisted prototyping. Then human review of design, copy and code. Only then deploy.
Code lives in a versioned repository, which makes it possible to roll back and lets someone else continue the project.
Conclusion
AI does not replace creative direction. It shortens the distance between deciding and seeing it work — and that's what moves a project forward.
Related questions
- Does AI replace developers?
- Do AI-built websites rank well on Google?
- How do you keep brand identity while using AI?
Written with AI support and reviewed by Douglas Pan before publishing.