Four phases every website is going through right now
Florian Berger · Jul 23, 2026
I believe we're only at the beginning of a shift that most people don't yet recognise as a shift at all. What we call a “website” — a bundle of designed pages for human eyes — is slowly fading into the background. What remains is something more sober: a connected, machine-readable data point. Here is how I see this development unfolding — four phases, from what is already everyday practice to what remains pure speculation.
Phase 1 — The web for humans (Yesterday)
The classic internet: built for eyes and for search engine crawlers, which essentially just count and link pages. The guiding principle was “more traffic on the site”. You typed a keyword, clicked through ten blue links, and pieced the answer together yourself.
Phase 2 — The hybrid web (Today)
This is where we currently stand. The internet has gained a second audience that must be taken just as seriously as the human one: language models. Ask a question today, and you increasingly get a ready-made, summarised answer, often without ever visiting a website. For an AI to give that answer correctly, it doesn't need a pretty design — it needs structured, machine-readable data.
To me, that's not an abstract vision of the future — it's exactly what we build into btlabs Core as standard: a foundation that feeds a website, a shop and AI agents simultaneously through open interfaces, with machine-readable profiles (`llms.txt`), so a language model can recommend a business without errors. Not as an add-on module, but as the baseline state.
Phase 3 — The acting web (Tomorrow, estimated 2–5 years)
Here I leave solid ground and speculate — reasoned speculation, but speculation nonetheless. AI stops merely reading things out and starts acting: negotiating, comparing, booking, without a screen ever being touched. Whether and how quickly this arrives, nobody really knows. That the technical building blocks for it are already emerging today, I see in every project.
Phase 4 — The distant picture (A pure thought experiment)
A picture I deliberately call what it is: speculation, not a forecast. Classic websites largely dissolve, companies talk about their own small language models for internal use, and because a large share of the web may by then consist of AI-generated content, provability becomes the real currency. Whether it turns out this way, I don't know. But the questions behind it — who an item of information truly belongs to — are already no longer science fiction today.
What this means in practice
You don't need to be ready for Phase 4 to act correctly today. Phase 2 is no longer optional — it is the baseline equipment on which every later phase is built.
Frequently asked questions.
How do I tell whether an agency genuinely understands how AI reads my website?
A simple test: ask about their last three projects and how AI visibility was handled. If the answer stops at meta tags, page speed and classic rankings, that only covers half the job — important, but not decisive for AI answer systems.
Anyone who genuinely understands how AI reads a website talks about more concrete things: a machine-readable profile such as `llms.txt`, structured data in JSON-LD format for facts like opening hours or services, and whether AI crawlers and agents actually have technical access to the site. These are the levers that decide whether an AI can cite or recommend your business correctly.
How do you measure success when AI answers no longer bring traffic to my site?
The classic click-through rate loses meaning once a user gets their answer directly from the AI, without ever visiting a site. The new metric can be called Share of Model Voice: how often, and in what context, is your business mentioned when someone asks ChatGPT, Perplexity or Claude a typical question from your industry — and is it described correctly and positively?
You measure this by regularly putting typical customer questions to the major AI systems yourself and logging whether and how your business appears. Success is then no longer shown by visitor numbers, but by being present and cited correctly, even when no direct click results from it.
Why doesn't good content for humans automatically translate into AI visibility?
Good content for humans is allowed to be entertaining, evocative and written in marketing language — “Your trusted partner for generations…” reads well to a human. An AI, however, can do little with phrases like that, because they contain no verifiable information.
What AI systems favour is information density: clear, self-contained statements with concrete facts, figures and relationships that make sense even without the rest of the page. The trick is combining both — marketing tone for humans, but with a factual core an AI can cite without having to interpret it.
How does a website build the trust an AI needs to prefer it over the competition?
AI systems orient themselves on signals similar to Google's: the E-E-A-T principle — Experience, Expertise, Authoritativeness, Trustworthiness. Translated, that means experience, expertise, reputation and trustworthiness must be recognisable, not just claimed.
In practice, that means: authors visible with real, verifiable qualifications rather than anonymous text, verifiable facts rather than vague promises, and content that shows genuine practical expertise. Text with no recognisable person behind it is harder for an AI to place than text with clear, provable authorship.
What is the difference between SEO, GEO and LLMO?
The three terms build on each other rather than replacing one another. SEO (Search Engine Optimisation) makes sure a page ranks well in classic search results. GEO (Generative Engine Optimisation) extends that goal to being cited as a source in AI-generated answers — in a ChatGPT or Perplexity answer, for example.
LLMO (Large Language Model Optimisation) goes a step further and gets more technical: it means specifically understanding and influencing how individual language models process, weigh and reproduce content — essentially reverse-engineering the model in question. In practice: SEO remains the entry ticket, GEO secures broad citability, and LLMO fine-tunes the details for the systems that matter most to a given business.
How do I get MCP access for my website, and what's included?
We set up MCP access for you whenever you want it — it's part of expanding your website, not a mandatory component. This lets you, or an AI assistant of your choice, work directly with your own website: maintaining content, querying your own data, automating recurring tasks. We agree together on what can be accessed — opening hours, services or contact details, for example — and you always keep an overview of who accesses what and when.
The effort is manageable: one-off setup, then the access runs alongside your website. It's most worthwhile once you update content regularly or have recurring questions about your own data — then an AI can take over the routine work, on request.
