What if a website adapted to you, instead of the other way around?

For years, websites have been designed around pages and navigation structures. As a visitor, you have to find your own way to the information or functionality you need. We turn that around.

A few months ago we wrote about the generative website and the Conversation Management System. With vragen.ai, we're now making that possible.

It started with a reliable, traceable answer based on your own information. Then came the visual elements, so an answer doesn't have to be text alone. Now we're taking another step: the page around the answer also adapts to what you (and so your visitor) need at that moment.

This is also called Generative UI. Take that development further, and how you use a website changes too.

From an answer to a page

Take a municipal website. Someone who is moving usually ends up on a page about moving. It brings together information for all kinds of situations: when you need to register your move, what you need, what happens to municipal taxes and where to go with questions.

With AI, we can now generate a good answer from that, and we don't have to stop there.

Say a visitor writes: “I'm moving to Deurne next month.” That makes the intent clear, and the website can first ask for the context it needs. Are you new to the municipality, or moving within Deurne? Are you moving alone or with a family? Do you have a car?

Based on that, a page can appear with exactly the parts that matter for this situation. A step-by-step plan for the move, a way to register it right away, information about waste collection in the new neighborhood, municipal taxes and, for example, parking if that applies.

Someone who is leaving Deurne gets a different page, from the same knowledge and the same website components.

Personalization without designing every variant in advance

Websites have been personalized for a long time. We show different content to new and existing customers, create separate customer journeys for target groups and adjust recommendations based on earlier behavior.

To do that, we usually have to come up with the possible variants in advance. We define segments, design the customer journey and set rules: if someone falls into target group A, we show variant B.

The more you want to personalize, the more variants you have to design, build and maintain.

With a generative website, that principle shifts. You design the building blocks the system uses to put together a fitting page. Mapping out every possible page or customer journey in advance is no longer necessary. You provide good knowledge, components, functionality, context and rules. Based on those, the system decides which combination helps at that moment.

That makes personalizing down to the last detail (hyper-personalization) suddenly feasible, without drawing a separate customer journey for every situation you can think of.

An example. Say you tell a travel website that you want to go to Ibiza for a week, like to spend time on the beach, want to visit the old town and are traveling with two children. Instead of ten blue links, a list of accommodations or a long AI answer, a page can appear with a map, suitable areas, a few accommodations, beaches, activities, reviews and a proposal for the trip.

If you then say you'd rather stay somewhere quieter, that same page can adapt to it.

It starts to look like what a good travel agent would do: first understand what you're looking for, then bring the options together. Only now the website puts those options together itself.

A generative website is about more than chatting

Will we all soon be operating our websites by talking to them?

A conversation is a good way to find out intent and context, but not the only one. On a municipal website, someone can just as easily click “I'm moving” and then make two or three choices. On a travel website, you can start with photos, destinations or preferences and only ask an open question later.

The input can be text, but also clicks, filters, forms, earlier choices or context the visitor has deliberately shared.

That makes ‘conversational’ broader than chat: two-way traffic, where the website adjusts what it shows with every click, choice or question.

So a generative website doesn't have to start with a prompt. It starts with the intent: what does someone want to get done? Then the system puts together the knowledge, actions and presentation that fit.

How does AI know what a website can do?

To build a website like that, access to the content isn't enough. The AI also needs to know which components and functions a website has.

That's what we're experimenting with in vragen.ai right now, using WebMCP.

Put simply, WebMCP lets a website tell an AI what's possible on that website. The website can, for example, make it known that there's a map component, a step-by-step plan, a form or a booking function. The AI can then use those options when it puts the experience together.

We built a first proof of concept for this with Drupal. Drupal provides the existing components a page can be built from. vragen.ai can use those components based on the visitor's question and context. It works the same way with other CMSs and frameworks.

You want to use components that designers and developers have made: tested, accessible and in your own brand style.

We then took the proof of concept further on our own website, which doesn't run on Drupal. That's where a live demo now runs.

You design the building blocks, not every single page

The expertise of designers matters just as much as before. You simply design in a different way.

A design system can become the toolbox for your AI. It holds more than colors, typography and buttons: complete components, such as a product card, step-by-step plan, comparison, map, form or call to action.

It also has to be clear how those components may be used. When does a comparison make sense? What information should a product card always include? Which combinations work and which don't?

The same goes for content. If information only makes sense within one carefully built page, it's hard to reuse dynamically. Content increasingly needs to be available as knowledge that can be used in different situations.

So you still design the website, just at a different level. The organization takes care of the knowledge, components, functionality and rules. Within that, the system decides which combination suits the visitor.

The next step: website and AI become one interaction

An AI assistant can know what functions a website has. But that doesn't mean the AI also understands what a visitor is seeing and doing on the website at that moment.

That's where the next step lies. Imagine the AI knows which pages someone has visited, where they are now and which filters or settings they've chosen. Maybe the AI even remembers what someone did during an earlier visit, if the visitor is fine with that.

With that information, the AI can respond much better to what a visitor is trying to do. The website and the AI assistant then form one experience together.

That changes the way we use websites. If someone clicks on a product, the AI knows where their interest lies. If someone asks a question, the website shows the right information or function right away. If someone makes a choice, the website adapts to it.

The result is a website where clicking, searching and asking questions naturally flow into each other. Visitors no longer have to choose between navigating the website and talking to an AI assistant.

From managing pages to managing possibilities

That also changes the role of the CMS. Today, a CMS is mainly set up to put pages together in advance. Content, templates and components come together in a fixed structure that is then shown to visitors.

In a generative website, those parts are still needed, but their final combination doesn't always have to be fixed in advance.

The CMS manages the knowledge and components, for example. A design system determines how they look and behave. Integrations make actions possible. The Conversation Management System uses those parts to put together a fitting experience, within the business rules you set there: what is and isn't allowed.

1. 1

   CMS

   Manages the knowledge and the components.

2. 2

   Design system

   Determines how those components look and behave.

3. 3

   Integrations

   Make actions possible, such as registering a move right away.

4. 4

   Conversation Management System

   This is where you set the business rules: what is and isn't allowed. Within those, it understands what the visitor wants to get done, picks the knowledge, components and actions to match and builds the page from them.

We build the ingredients that different experiences can grow from.

One visitor wants to arrange a move quickly. Another wants inspiration for a vacation. One types a detailed question, the other would rather click three times.

The website doesn't need to know all those routes in advance. It does need good knowledge, well-designed components, reliable functionality and clear rules.

For web teams, the question shifts from ‘which page should we build?’ to ‘what do we need to make available to help every visitor well at that moment?’