There’s a strange experience becoming increasingly common with AI-built websites.
You ask ChatGPT to help you build a site. It writes the copy, suggests the structure, designs sections, generates code, and helps you make revisions. The result looks pretty good.
Then you ask a different question:
“Is this actually a good website?”
Suddenly, the same AI that helped create it starts finding problems.
The positioning could be clearer. The page feels generic. There’s too much repetition. The visual hierarchy is weak. The proof points aren’t strong enough. The design looks like other AI-generated websites.
So why didn’t it fix those things while it was building the site?
Because generating something and judging whether it is good are different tasks.
AI is very good at giving you what you ask for
Most AI website workflows are iterative. You make a reasonable request, get a reasonable result, and then keep building from there.
Each individual request can produce a perfectly reasonable result.
The problem is that a great website isn’t simply the sum of a series of reasonable requests. It needs an underlying strategy.
Those decisions are easy to lose when a website is assembled one prompt at a time.
Creation and evaluation put AI into different modes
When you ask AI to create something, the primary objective is usually to satisfy the request.
When you ask it to critique something, the objective changes. Now it is looking for weaknesses.
An AI can comply with your request to add another three-column card section and later correctly observe that the page has too many repetitive card layouts.
It can write a headline you approve and later tell you that the positioning isn’t differentiated enough.
It can make every section independently attractive and later notice that the overall page lacks rhythm.
We think of this as the generation–evaluation gap: the difference between what an AI system is willing to generate and what it recognizes as high quality when explicitly asked to evaluate the result.
A website is a system, not a collection of sections
AI is often remarkably good at local optimization. Give it one section and it can improve that section. But improving one section does not necessarily improve the website.
Make every headline larger and eventually nothing feels important. Make every section more visually interesting and the page becomes exhausting. Add more explanation everywhere and the story gets harder to understand. Give every section a unique layout and the site loses consistency.
Good design requires balancing local decisions against the whole. That’s why an experienced designer will sometimes make an individual element less interesting because it makes the overall composition better.
The missing step is judgment
The breakthrough in generative AI has made creating things dramatically easier. But creation was never the entire job.
Good creative work also requires deciding what should be created, what matters most, what should be removed, what should remain consistent, when consistency should be broken, whether the result actually communicates what it needs to, and whether the result is good enough to ship.
Those are judgment problems. And they become more important as generation gets easier.
How to get better results
If you’re building a website with ChatGPT, Claude, or another general-purpose AI tool, don’t use it only as a generator. Separate the process into stages.
Ask whether the page is accomplishing its purpose, not simply whether it looks good. Ask what feels generic, what information is missing, whether the hierarchy is clear, whether the design has drifted from earlier decisions, and what should be removed.
AI becomes much more useful when it isn’t responsible only for producing the next thing.
This is a bigger problem than website generation
At Morphic, this question has shaped how we think about creative AI more broadly.
The future isn’t simply better generation. A useful creative system needs to understand, plan, create, evaluate, refine, and remember.
It needs to retain the decisions that led to good work so the next thing doesn’t start from zero.
That’s the difference between an AI that can make things and a system that can help a company consistently make good things.