We have been receiving more and more websites that clients built from start to finish with artificial intelligence, from the design to the legal texts. To some extent we understand why. Today it is possible to have a site online in an afternoon, at almost no cost and with a look that, at first glance, seems professional. We also use AI every day at Multisnet and have no intention of stopping.
Until now, when we looked at these sites, our comments were almost always about design, because they all looked alike, and about SEO. We have already written about this regarding copywriting and, honestly, we thought the subject was more or less settled. These are problems that cost visibility and customers, but rarely go beyond that.
This week we came across a case that made us look at the subject differently, because this time the problem was in the legal texts.
A client of ours had built and published the website for their new business on their own. At first glance everything was there, including a privacy policy with sixteen sections and references to the GDPR and the Portuguese data protection authority. It was only when we read it more carefully that we realised that what had been published was, in fact, a draft.
When you ask an AI tool to write a privacy policy, it usually leaves notes about what still needs to be confirmed. The problem is that it writes them in the same tone as the rest of the document and, if nobody catches them, they end up published as if they were final text. In this case we found several, which we summarise here in our own words:
The most delicate point was the cameras. The company wanted to have them in the rooms and the policy explained that, before switching them on, it was necessary to confirm whether the law allowed it. In practice, a public document, which any customer or the data protection authority itself can read, in which the company admits it does not yet know whether what it wants to do is legal.
The terms and conditions were no different. One of the points said that the privacy policy should be made available through a visible link on the website, when that link already existed and the policy was already published.
What worried us most was realising that, for anyone who is not used to reading these documents, the text looks flawless. It is well organised and cites the right legislation, so it is natural that nobody suspected anything.
The tool even warned about what was left to do. Those warnings simply stayed in the middle of the text and were published with it.
After the privacy policy we looked at the rest of the site and found another pattern. Each page seemed to have been generated in a different conversation and nobody had read the whole thing in one go:
None of these flaws is serious on its own, but together they give the impression of a business that does not yet quite know how it works. In the case of the ages, it is the kind of thing that ends in an argument at the front desk.
Then there is the problem we already knew and that, frankly, has bothered us for quite some time. AI-generated sites are becoming so alike that we can often recognise them within the first few seconds. The structure is almost always the same:
The colour and the logo change, but the structure is the same, whether it is a dental clinic or a gym. It seems to us that this happens because AI tends to return the most likely answer, which ends up being an average of everything it has seen. For a technical text that may even be useful, but a brand's identity needs precisely the opposite.
This site also had a curious detail, a selector with eleven colour themes, including one for Halloween and another for Christmas. Nobody had asked for it and, meanwhile, the cookie panel and the booking system were still unfinished.
And what you cannot see? No page on this site had a description for search engines and, on the homepage, the names of the activities were only inside images, with no alternative text. For Google, and for anyone browsing with a screen reader, the page did not explain what you can do at the venue.
Not at all. We use it and will keep using it, but there is always someone who knows the project reviewing the result before it reaches the client. For those who have already built their site with AI, here are a few checks that would have caught almost everything we found in this case:
This case shows how a tool used with good intentions can create a serious problem when nobody reviews the result. The AI wrote down what was left to do, in plain language, but on the other side there was nobody who knew what to look for before clicking "publish".
We still believe AI has a place in a web project and, in our work, it has an ever larger one.
A website is the face of a business and, in its legal texts, it is also a commitment to customers and to the law. That is why, before it goes live, it should always pass through the hands of someone who knows the subject.
I want a second opinion on my website
Do you know the difference between these two techniques that are part of a solid SEO strategy?
A global cache rule on Cloudflare caused logged-in customers to see other users' names and carts. Here is what happened and how we fixed it without losing performance.
