Optimising your website for AI search engines: how business websites are really built in 2026

Imagine a customer no longer asks Google but ChatGPT for a tax adviser in your town. What happens to your website then? Does it show up in the answer at all, or is it simply skipped?

That is exactly what is happening right now, without any big announcement. Alongside classic search, AI systems such as ChatGPT, Perplexity or Gemini now read websites too and summarise them for their users. Anyone who builds only for Google is building past reality. This article shows what has changed in the technical side of building websites and what "readable for AI" means in practice for your company.

What has really changed in building websites in recent years

A few years ago, the process for a new company website almost always went the same way: brief an agency, wait for weeks, a CMS with a dozen plugins, live at some point. Today, much of that looks different.

The building process itself has become faster because AI tools help with programming and writing. That is no longer a secret but everyday practice in good development teams.

What has actually changed:

• The building process itself often takes days instead of weeks, because routine tasks run automatically
• Content, structure and technology are planned together from the start, not one after another
• Quality assurance becomes more important, not less, because building faster can also go wrong faster
• Websites are less often a collection of plugins and more often a lean, purpose-built system

That does not mean everything suddenly runs by itself. A human still decides what goes on the site, what it looks like and whether a text really fits the company. Only the tools in between have changed.

Building faster does not automatically mean building worse

Sounds risky at first, doesn't it? More speed, less control would be the obvious worry.

But the point is: speed and care are not mutually exclusive. The real question is what happens between "code written" and "site live".

An example: a trade business has its new website built. AI-assisted tools deliver a working first version in a short time.

Before any of it becomes publicly visible, everything goes through a staging environment. There, a human checks every page visually, and automated tests check forms, links and loading times. Only then does the site go live.

Why "optimised for Google" is no longer enough

For years, classic SEO was the only target: good rankings on Google, done. That goal remains important, but it is no longer the only one.

More and more people put their questions directly to an AI assistant instead of a search engine. "Which provider in my region does X?" or "Roughly what does Y cost?" Questions like these increasingly end up in a chat window instead of the Google search bar.

The technical term for this is generative engine optimisation, often called GEO. It means preparing your content so that an AI system understands it, classifies it correctly and, at best, quotes it. Anyone missing entirely is simply not mentioned in that answer.

Does the term still sound unfamiliar? The principle behind it is simple all the same. You no longer write only for an algorithm that counts keywords, but for a system that tries to actually understand your content.

An everyday example: anyone who asks an AI assistant for a dentist in a particular town with experience treating anxious patients often already gets a concrete answer with names today. Not ten blue links to click through yourself, but a direct recommendation.

What AI search systems actually do with your website

AI assistants read websites differently from a person scrolling. They break a page into blocks of text, try to recognise connections and look for clear statements they can quote.

A page full of vague marketing language offers little that can be quoted. A page with clear sentences like "We build websites for companies in Germany, prices from X euros", on the other hand, gives the system something tangible.

This is not about forcibly bending texts for machines. Clear, honest sentences about your own offer read just as well for people as for an AI system, often even better.

An example: the question about the painting company

Someone asks an AI assistant for a painting company in the region with experience in listed buildings. The assistant searches the available sources and prefers pages that state clearly which services they cover, where they are based and what they specialise in. A home page with nothing but a slogan and three stock photos has poor chances here.

Structured data: the business card for machines

Structured data is an additional data set in the background of a page, invisible to people. It tells search engines and AI systems explicitly: this is a company, this is its name, its address, its offer, its opening hours.

Without this data, a machine has to guess what a page is about. With it, the machine does not have to guess; it simply reads it off.

Schema.org: the shared standard behind it

The technical term for this vocabulary is Schema.org. Google, Microsoft, Yahoo and Yandex launched the project together in 2011 so that not every search engine would invent its own language for "this is a company" or "this is a product". Since then, Schema.org has been the common denominator that practically every system reading websites automatically relies on, including today's AI assistants.

In practice, it usually looks like this: a piece of code in JSON-LD format, invisible to visitors, delivered in the background of the page. Only a few Schema.org types are relevant for a company: LocalBusiness for address and opening hours, Organization for company details, FAQPage for question-and-answer sections, Product and Review for offers and ratings, Article for blog posts like this one.

You do not need to know these types by heart. It is only important to know that they exist, and that a good website uses them correctly in the background instead of mentioning the company name, address and offer somewhere in the running text.

What structured data achieves in practice:

• Search engines often show additional information directly, such as rating stars or opening hours
• FAQ sections can be recognised and quoted as clear question-and-answer pairs
• AI systems can clearly assign company, location and services instead of interpreting them
• The likelihood increases that your site appears as a source in an AI answer

A trade business with properly stored structured data for address, services and customer reviews has a clear advantage here over a website where this information only appears somewhere in a running text on the page.

Take a restaurant that has its opening hours only as an image in the footer. Neither Google nor an AI assistant can reliably read from it whether it is still open tonight. With structured data, this information is available in machine-readable form, regardless of the image next to it.

llms.txt: a sitemap for AI language models

Most websites have a robots.txt and a sitemap.xml. One tells search engine crawlers what they may visit, the other lists all pages clearly.

An llms.txt does basically the same, only for AI language models. It is a simple text file in the website's root directory that summarises in short, clear form what the site is about, which areas are important and where the relevant content is.

The format was proposed in 2024 by Jeremy Howard and his team at Answer.AI. There is deliberately no official standard with fixed rules, rather a convention: a Markdown file at the address /llms.txt that starts with a short H1 heading, a summary in one or two sentences and then links to the most important subpages with a short description. Some websites additionally create an llms-full.txt that provides the complete content as a single long text file.

To be completely honest: not every AI system actively reads this file yet; support is only just growing. Anyone who sets it up anyway loses nothing. It costs hardly anything when the website is built, and those who are early do not have to retrofit later.

Instead of an AI system fighting its way through navigation, cookie banners and ad spaces, it immediately gets a clean overview. For a medium-sized company, that means an AI assistant understands in seconds what the company does, instead of clicking through ten subpages and missing half of it.

How to tell whether your website is readable for AI

You do not need to become a developer for this. A few simple questions are enough for a first check.

• Does the home page say in one or two clear sentences what your company does and for whom?
• Is there Schema.org data (e.g. LocalBusiness or Organization) in the background of the page?
• Is there an llms.txt file that gives AI systems a summary of your website?
• Does at least one page answer typical customer questions in clear question-and-answer form?
• Does the page load fast enough to be read in full at all?

If you answer no to more than two points, a closer look at your own website is worthwhile.

"Only looks finished" vs. "actually works"

This is the catch with many quickly built websites: at first glance they look finished. A colourful home page, a slick animation, everything seems professional.

Only after a few months does the difference show. Forms that silently stop sending emails. Loading times that slowly creep up. Structured data that was never set up properly because there was no time for it in the rush.

A website that actually works differs in what you do not see straight away: a solid technical foundation that still loads quickly in a year, is secure and can be extended without having to be rebuilt completely.

Human quality control remains the difference

AI tools make building a website faster. They do not automatically make it right.

That is exactly why a review step before every launch remains important, in which a human actually looks at the site, clicks through it and tests it.

What good quality control before launch covers:

1. Visual check of every page on all common screen sizes
2. Automated tests for forms, links and loading times
3. Checking the structured data for completeness and correctness
4. A final look at spelling, image rights and privacy information
5. A staging review before anything becomes publicly visible

In conversations with clients we notice again and again that exactly this step is often missing when a website was put together at top speed with a single prompt. The result looks finished. Whether it really works only becomes clear later, and then usually at the most inconvenient moment.

German hosting and GDPR as the foundation, not an extra

One point that easily gets lost in the discussion about AI and website building: where the data is stored and how it is protected remains just as important as before.

For a company with customers in Germany, German hosting and a GDPR-compliant setup really belong to the basic equipment. Not an add-on module you can book later, but part of the foundation from the start.

A website that impresses technically, but whose contact-form data ends up somewhere outside the EU, creates more problems in the end than it solves.

What this already means for your company

Perhaps you are thinking: this concerns large corporations or tech companies rather than my medium-sized business. Is that really true?

Why this is not a niche topic for tech corporations

A regional service provider, a trade business, a law firm, a shop with an online store: in all these cases, people search for exactly these offers, and a growing share of them use an AI assistant instead of classic search to do so. Anyone who is already set up to be readable for AI today will have a head start tomorrow that cannot easily be caught up.

The number of enquiries that reach a company via an AI assistant is still modest; there is no denying that. But websites that have to retrofit the technology need time, a budget and often a real overhaul rather than a small adjustment for it. Anyone who considers the topic now saves themselves that stress in a year or two.

How do you imagine a potential customer will come across your company a year from now? Through a Google ad, or through the answer of an AI assistant that names your website as a trustworthy source?

How to approach the topic without becoming a developer yourself

The good news: you do not need to understand how structured data is technically programmed. You only need to know what to look out for when you plan a new website or have an existing one reworked.

1. Ask your agency or development team directly whether Schema.org data and an llms.txt file are part of the offer
2. Check whether there is a real test and review step before launch, not just a quick "looks good"
3. Make sure your most important services and locations are on the site in clear, simple sentences
4. Clarify where your data is hosted and whether GDPR compliance is considered from the start
5. Plan for AI visibility from the beginning instead of tackling it later as a retrofit

What counts in the end

Search is changing noticeably right now, and with it the requirements for a company website. A site built only for Google already no longer covers everything that counts.

What does your own website look like when you read it through the eyes of an AI assistant that has to decide within five seconds whether it is worth an answer?

But Right builds websites according to exactly these principles: with a modern, proven tech stack, Schema.org data and an llms.txt from the start, and with a real human review before every launch instead of just prompt-and-done. If you are planning a new website and want to build on a solid, AI-readable foundation right away, you will find more at websites.but-right.com/en/new-website.