"Can't the AI just do that?" I now hear this question in almost every first meeting. And it's a fair one: anyone who needs a website today reads everywhere that artificial intelligence writes texts, creates images and even programs. How much of that is true, and where are the limits? In this article I give you an honest assessment – from the point of view of someone who works with AI tools every day.
Which AI tools exist today
Roughly speaking, the offerings fall into three groups:
- Website builders with AI: Many website builders create a complete first draft with page structure, texts and images after a few questions about your industry and offer. You then continue editing in the builder.
- Chat assistants: Tools like ChatGPT, Claude or Gemini help with writing, translating, summarising or creating images. They don't build a website, but they provide a lot of material for one.
- Coding agents: Tools like Claude Code work directly in a website's source code. They read the existing code, suggest changes and implement them. They are steered by a developer who sets the goals, makes the decisions and checks every result.
I mainly use the third group. The new version of this website was built that way too – I describe how that works in practice in the article How I work with Claude Code.
Where AI really helps
I don't want to play this down: in some areas the gain is huge.
- Speed: Routine work such as creating similar pages, reworking templates or updating dependencies is much faster.
- First drafts: Whether page structure, text outline or design idea – a first proposal is ready within minutes. That makes it easier to talk about something concrete instead of a blank page.
- Texts: AI helps with revising, shortening and making texts consistent. Especially if you don't enjoy writing, it gets you to a usable starting point more easily.
- Translations: A second language is no longer a big project. A review by someone who knows the language well still makes sense, though.
- Routine code: An agent handles recurring programming tasks reliably, as long as it is clear what the result should be.
An example from my own work: I added three new languages to my web app Shared Lists – it took no longer than an hour.
How much faster is it really?
The research on this is surprisingly mixed. A large field study with almost 5,000 developers at Microsoft, Accenture and another large company found around 26 percent more completed tasks with an AI assistant (Cui et al., Management Science). A smaller study by the research organisation METR, on the other hand, measured that experienced developers took 19 percent longer with AI in their own, very large open-source projects – even though they were convinced they were faster (METR, 2025).
Both studies, however, look at tools that are already outdated: mainly code autocompletion and tools as they were in early 2025. Coding agents that implement whole tasks across many files on their own have developed a lot since then.
My own experience with websites and web applications is clear: the time needed for implementation has at least halved. Since I bill by effort, the price for you can almost halve as well. On top of that comes a share of the cost of the AI tools – but compared with the working time saved, it is small.
Where it gets difficult without expertise
An AI quickly delivers something that looks good. Whether it actually is good is not something it decides. This is exactly where the work begins that should still be done by a person with experience.
Clarifying requirements
The most important question for a website is not "What does it look like?" but "Who should it reach, and what should those people do?". An AI answers the questions you ask it. Asking the right questions remains your job and mine.
Data protection and law
A business website in Germany needs a legal notice (Impressum) under § 5 of the Digital Services Act (DDG, the successor of the Telemedia Act) and must comply with the GDPR. In addition, under § 25 TDDDG, information may only be stored on or read from visitors' devices – for example via cookies – if this is strictly necessary or if consent has been given. And personal data such as the IP address should not be passed on to third parties without asking, for example through externally embedded fonts, videos or analytics tools.
AI-generated pages and website builders often embed such services by default. Recognising which of them are needed and which are not requires expertise. (This is an assessment from practice, not legal advice.)
Accessibility
The German Accessibility Strengthening Act (BFSG) has applied since 28 June 2025. It covers certain digital services for consumers, for example online shops or online appointment booking. Micro-enterprises – fewer than ten employees and an annual turnover or balance sheet total of no more than 2 million euros – are exempt for services. Whether your own offer is affected can be checked at the Federal Accessibility Agency (German). The technical benchmark is the European standard EN 301 549, whose new version V4.1.1 from September 2026 builds on the Web Content Accessibility Guidelines (WCAG) 2.2 at level AA.
But regardless of the law, every website should be usable by everyone: sufficient contrast, meaningful headings, alternative texts for images, keyboard operability. AI output sometimes meets these requirements and sometimes doesn't – it always has to be checked.
Security, hosting and maintenance
A website isn't finished once it goes live. Software has to be updated, certificates renewed, backups made. AI can help with that, but it doesn't feel responsible when something fails at night. How quickly an outdated website becomes a problem is something I experienced in a client project: there, a CMS that could no longer be updated eventually led to a complete outage.
Quality control
AI tools make mistakes – and they make them with great confidence. An agent can write code that works but is unnecessarily complicated, contains security holes or breaks something elsewhere. Texts can be factually wrong or sound like every other website. That's why my rule is: no change goes online that I haven't read and understood myself.
Responsibility
In the end, someone has to vouch for the website working, being legally sound and fitting you. No software can take that on.
When is an AI website builder enough?
Honestly: sometimes. If you need a simple business card website, like working on it yourself and accept the limits of the builder, it can be a good start. In that case, pay particular attention to where the data is stored, which services are embedded and whether you can take your content with you later.
As soon as more is involved – several people maintain content, there are forms, bookings or member areas, or the website is an important part of your business – professional support pays off. And if you're unsure, let's simply talk about it. You can also get an independent assessment from me if you end up implementing the project differently.
Conclusion
AI has noticeably changed the work on websites. I have become faster, and my clients benefit from that. So to the question "Can't the AI just do that?" I usually answer: yes, a lot of it – but someone has to know what it should do. The focus has shifted from "doing" to "being creative": implementation no longer takes most of the time; the question of what a website can make possible does.
The tools don't replace the knowledge of what a good, secure and legally sound website looks like, though. Someone has to ask the right questions, check the results and take responsibility.
If you're thinking about how to approach your website, with or without AI, feel free to write to me. I'll tell you openly what makes sense in your case.
Yours, Michael Becker