DRAVVT

Websites · Artificial Intelligence

Building your website with AI: what it does well and the mistakes it makes on its own

Today you can have a website in ten minutes with Wix, Canva or an AI. I use AI every day for my own work. But there’s a gap between having a website and having one that brings you customers, and AI doesn’t cross that gap on its own. Here’s what it does well, where it fails, and how to use it without it costing you.

By Dravvt · October 2026 · 9 min read

A robotic hand and a human hand about to touch the letters “AI”, with the title “Building your website with AI: what it does well and the mistakes it makes on its own” and the dravvt logo

What AI does well

Let’s start by giving it credit: artificial intelligence has changed the way websites are made. You describe your business and within minutes you have a design, some starter text and photos that fit. For someone who has never had a website, that’s a huge leap.

I use it for what it does best: writing a first draft that I then fix, reviewing code, testing ideas quickly and saving hours on repetitive tasks. If you want to see how these tools work under the hood, in this article I compare the AI coding agents I use every day.

The problem isn’t AI. The problem is leaving it alone.

The mistakes it makes when it works alone

The letters “AI” in relief on a circuit board, with the title “What AI doesn’t do on its own” and the dravvt logo

1. It makes things up and tells you with complete confidence

Language models, the technology behind ChatGPT and the like, sometimes guess when they don’t know, and they do it with the same confidence as when they say something true. In 2025 a group of OpenAI researchers published a study on why this happens, and they open with examples that are funny and scary at the same time. They asked the AI DeepSeek for the birthday of one of the authors, telling it to answer only if it knew: it gave three different dates, and all three were wrong. Then they asked ChatGPT, DeepSeek and Llama for the title of that same author’s doctoral thesis: each one gave a different title, university and year, and none of them got it right (Kalai et al., 2025).

Their explanation is easy to grasp with a multiple-choice exam. If a blank answer is worth zero and a guess might be worth a point, a student is always better off guessing. AI is in the same position: it’s trained and evaluated with tests that reward right answers and give nothing for saying “I don’t know”, so it learns that its best move is to answer confidently even when it has no idea (Kalai et al., 2025).

On a website, that turns into opening hours that aren’t yours, a service you don’t offer, a price pulled out of thin air or the wrong address. And if nobody checks it, that false detail is the first thing your customer sees.

I’ve lived this many times working with AI. Without supervision, it leads you down false paths: it tells you lies convinced they’re true, and if you don’t already know better, you believe them, because it talks to you like an expert. In the end you do the work twice, or you simply never get where you wanted to go.

2. It leaves security holes

Writing code that works is not the same as writing code that’s secure. Veracode, an IT security company, put more than 100 AI models to the test: it asked them to write code for real tasks and then ran it through its security tests. 45% of what they wrote failed those tests, with some of the best-known and most exploited flaws on the internet, and newer models did no better (Wessling, 2025).

This isn’t theory. In 2025 it came out that many sites built with Lovable, one of the fastest-growing tools for building websites with AI, left their database open: anyone, without logging in, could read or change the stored information. The flaw was recorded in the official US vulnerability database as critical, scoring 9.3 out of 10 (National Institute of Standards and Technology [NIST], 2025). If your site stores customer data, bookings or payments, that’s a risk for them and for you.

3. It doesn’t know who it’s talking to

AI doesn’t know what your customers type into Google when they look for you, who your competitors are in your city, or why people choose you over the business next door. It writes correct, generic text, the same text it writes for thousands of similar businesses.

And Google notices. Google doesn’t penalize you for using AI: it says clearly that using it appropriately isn’t against its rules, as long as the content is helpful to people (Sullivan & Nelson, 2023). What it does penalize is filling the web with mass-produced pages that add nothing (Google, 2026b), and among its warning signs it lists exactly that: using extensive automation to produce content on many topics (Google, 2026a). What Google rewards is the opposite: content that shows real experience and builds trust.

4. It doesn’t get the site ready to show up on Google

A good-looking website isn’t a website that shows up on Google. That takes things you can’t see: a title and description written for what your customers search for, a page for each service, structured data so Google understands what you do and where, speed on mobile and a Google Maps listing connected to the site. Google’s own AI reads all of this too when it recommends businesses; I explain how these assistants work in what AI agents are.

AI can do some of these things if someone tells it exactly which ones and then checks the result. On its own, it almost never does them.

5. It has no direction

This is the mistake that’s hardest to see and the one that costs the most. A website isn’t a brochure: it has to achieve something. Getting people to message you, book, or call. That means deciding what comes first, what to say and what to leave out, who it’s for and what visitors should do next.

AI doesn’t make those decisions for you. Something I always notice: if you don’t know where to go, AI won’t tell you. It gives you an answer, but it doesn’t give you experience. It answers what you ask, and often what you’re looking for isn’t what you need. The result is a nice-looking site that nobody finds and that, if someone does find it, doesn’t tell them what to do.

Even professionals need to check AI’s work

You might think this is just about using it badly, and that in expert hands AI already works on its own. The data says otherwise.

  • In the 2025 Stack Overflow survey, the largest survey of developers, 46% said they distrust the accuracy of AI and only 3% said they highly trust it (Stack Overflow, 2025).
  • METR, an institute that measures what AI can really do, ran an experiment with experienced developers working on their own projects: some tasks were done with AI and others without it, assigned at random. With AI they took 19% longer, yet afterwards they believed they had been 20% faster (Becker et al., 2025). Checking and fixing AI’s work takes time too.
  • In security testing of websites, AutoPT, an AI agent designed to find flaws in websites on its own, completed only 41% of its tasks (Wu et al., 2025). Another system, which splits the work among several specialized agents, cut its made-up errors from 12% to 5%, but still completed only 68% of its tasks (Singh Waldia et al., 2025).

AI is moving very fast, and every year it does more. But today, even in specialists’ hands, it needs someone to direct it and check its work.

What about free websites from Wix or Canva?

They’re a good way to start if you have nothing. But it’s worth knowing what comes with them:

  • The address isn’t yours. On Wix’s free plan your site lives at an address like yourbusiness.wixsite.com/yoursite and shows Wix ads on every page (Wix, n.d.-a; Wix, n.d.-b). On Canva it sits on a canva.site subdomain, with a “Designed with Canva” notice at the bottom (Canva, n.d.).
  • It’s not easy to move elsewhere. If you ever want to switch, you usually have to start from scratch.
  • The SEO is up to you. The template doesn’t know what your customers search for.

For a business that wants to be found, having your own domain (yourbusiness.com) is part of building trust.

AI as an assistant, not the boss

AI is great to work alongside, but someone has to be at the wheel. Building a website with AI that actually works means checking every detail, taking care of security, knowing what your customers search for on Google and deciding what the site needs to achieve. That’s what AI doesn’t give you: experience. And it’s exactly the work nobody sees when someone tells you a website takes ten minutes.

Frequently asked questions

What is the best AI to build a website?
The tool matters less than it seems. All of them, from the simplest to the most advanced, share the same limits: they can make things up, leave security holes and don’t know what your customers are looking for. The difference isn’t the AI, it’s who directs it and checks its work.
What kind of website can I build with AI?
A simple introduction page works fine: who you are, what you do, photos and how to contact you. The more the site does, the more careful you need to be. If it will store customer data, take bookings or handle payments, a security flaw can expose that information, as happened with many sites built with Lovable in 2025 (NIST, 2025).
Can AI build a website?
Yes, in a few minutes. What it doesn’t do on its own is check that the facts are true, make it secure, plan it so it shows up on Google or decide what it needs to achieve. A website that exists is not the same as a website that brings you customers.
How do you use ChatGPT to build a website?
ChatGPT can write the text and even the code for a website, but it doesn’t publish it, doesn’t connect it to your Google listing, doesn’t make it ready to show up in search, and it can make things up with complete confidence (Kalai et al., 2025). For any of that to be useful, someone has to check every line, know what’s missing and get it live. If you don’t have time for that, or don’t know what to check, that’s where a professional comes in.

How I do it

I work from data. Data can achieve a lot: it tells you far more than ten AI agents put together. Data reveals who your competitors are, where you stand today and what your customers really search for on Google. With that I decide what your website needs to achieve, and I check every line before it goes live. That’s how I built, for example, the website of Dr. Mauricio Herrera, a urologist in Colombia: urologoyopal.com.

If you want a website that people find and that brings you customers, here’s how I work.

Keep reading

References

We use cookies to improve your experience on this site. Learn more