We tested AI (ChatGPT) to code a website in 1 hour here is the surprising result

Author: Boisdur-Salbris Matthieu

For this test, we started from a real client brief: a small showcase website for a local services business, with a home page, an about page, a contact page and a form. The constraint: one hour of actual work, timer running. "Done" meant producing a first working version, readable on mobile and ready to show the client, not a perfect site shipped to production. We used ChatGPT to generate the code, the copy and the structure, then noted what went well and what still required human input.

What AI does really well

Generating a clean HTML/CSS boilerplate. Asking for a full page skeleton, header, navigation, sections, footer, with a basic reset and colour variables is almost instant. The result still needs a review, but the base is solid, consistent and well organised.

Producing section skeletons quickly. A prompt like "a hero section with a headline, a subheading and a button" returns a block that is ready to drop in, with correct semantic markup. You then assemble the blocks like pieces.

Drafting copy variations. Getting three phrasings for a headline or an about paragraph takes seconds. It gives you something to refine instead of a blank page, and helps you break out of your own wording.

Writing documentation. Generating a README, commenting the code or summarising the technical choices at the end of a session avoids spending time on a task that rarely feels urgent.

Iterating on small, isolated components. For a button, a card or a form, chaining small tweaks ("make it more accessible", "add a loading state") is fast and comfortable.

What AI cannot replace

Requirements gathering and business scoping. Before writing a single line of code, you need to understand the client's business, goals, audience and constraints. AI answers what you ask; it does not ask the right questions and does not surface an unspoken need.

UX trade-offs and accessibility. Choosing an information hierarchy, deciding on a navigation flow, checking contrast, keyboard navigation or alt text are design decisions the tool will not validate for you.

Architectural decisions. Which hosting, which framework, how to structure the project so it stays maintainable and can grow, these choices commit the project's future and require a wider view.

Security hardening. Protecting a form, validating data server-side, managing access or preventing injections: AI can suggest ideas, but it cannot guarantee safe code in a real production context.

Testing and the responsibility of going live. Checking that everything works across browsers and devices, anticipating failures and monitoring the site, a human takes responsibility at release time.

Long-term maintainability. Code generated without a view of how the project will evolve can become hard to extend. Documenting, structuring and keeping everything coherent remains a developer's job.

The real result

The verdict is balanced. Yes, AI saves real time on execution: what used to take half a day to set up can be prepared in minutes. No, it does not remove the developer's role, that role shifts towards orchestration, quality control and architecture. In one hour, AI produced most of the structure and copy; it was the scoping decisions, accessibility fixes, security checks and overall coherence that made the difference between a fast draft and a site ready to show.

Three habits to keep when using AI on a real project:

1. Give a precise brief. The more detailed the request, goal, audience, constraints, tone, the more useful the output. A vague prompt produces vague code.

2. Break the work into small steps. Generate one component at a time rather than an entire site, and validate each block before moving on.

3. Review and test everything. Treat AI output as a draft: check the code, test on mobile and keep control of security and deployment.

Conclusion: AI is an excellent co-pilot, not an autopilot.