Your AI Built the Site. Did Anyone Review the UX?

I like AI builders. I use them, and I am still a little surprised by what they can do. A person can sit down after dinner with a rough idea, a half-formed prompt, and a cup of coffee, and before the coffee is cold there is a page on the screen that looks as if a small team had been at work. The buttons line up. The sections have names. The page has a headline, a little proof, a gradient perhaps, and the general air of something finished.

That is the danger. Speed can make unfinished work look complete. A page can be built before it has been understood. It can pass the first test, which is whether it exists, and fail the second, which is whether a stranger can make sense of it without being coached.

The problem is not that AI makes ugly pages. Often it does not. The problem is that AI is very good at producing the shape of a page and less reliable at producing the judgment behind one. It can write a headline that sounds right and still says almost nothing. It can make a feature section look balanced and still bury the one thing the buyer needed to know. It can give me a page that is correct in the way a room is correct when the furniture is against the walls, while missing the fact that no one wants to sit there.

Research gives this problem some weight. The OpenAlex record for “Attention web designers: You have 50 milliseconds to make a good first impression!” points to work showing how quickly people judge visual appeal. Another OpenAlex-indexed paper on visual complexity and colorfulness studied 450 websites and 548 people, and found that those visual features, along with age and education, helped explain first impressions after only 500 milliseconds. That is not enough time for the visitor to appreciate my information architecture. It is barely enough time for the page to say, by order and proportion, “you are in the right place.”

Google says the same thing in plainer, more practical language. Its Search Central guidance asks creators to make “people-first content”, and its SEO starter guide keeps returning to usefulness, clear organization, and readable text. These are not glamorous ideas, but they are the ideas that matter. If a page is hard for a person, it is unlikely to become easier merely because the code is tidy.

This is where I think AI-built sites need a second pass. Not a pass for syntax, and not only a pass for speed, but a pass for human sense. When I look at one of these pages, I try to imagine the visitor arriving with no background and no obligation to be generous. The visitor has not watched me build the product. The visitor has not read my notes. The visitor sees the first screen and asks what the thing is, whether it is for them, whether it can be trusted, and what happens if they click.

Those questions are small, but they are severe. A vague answer at the top of the page makes everything below it work harder. A generic button makes the next step feel uncertain. A missing proof point turns a simple claim into an errand the visitor must run for me. The page may still be attractive, and it may still be fast, but it has begun to spend the visitor’s patience.

A useful review therefore has to ask who the visitor is. A developer may enjoy a dense feature list. A CFO may read the same section and wonder about risk. A busy parent may want the price, the return policy, and the shortest path through the page. A first-time visitor may need the whole offer explained in one plain sentence. None of these people is wrong. They are simply carrying different questions.

This is why I do not treat “AI built it” as the finish line. I treat it as the first draft. The machine can assemble the page, and that is useful. But before I trust the page with traffic, I want to know what it feels like to the people it is supposed to persuade. If several different personas stumble over the same sentence, I do not defend the sentence. I rewrite it.

AI can build the site. I still have to make sure a human can read it.