I. The graph everyone reads wrong
There’s a graph showing up on a lot of marketing teams’ desks lately. The one tracking organic traffic. It’s dropping. Not off a cliff — but steadily, measurably, across sectors.
The conclusion that quickly follows: “the website is losing relevance.” Some teams start thinking out loud about a smaller site. Fewer pages. Less maintenance. Budget redirected to where the people still are — social, paid, in-app.
That’s one of the most dangerous conclusions you can draw today.
What that graph measures isn’t the relevance of your website. It’s the behaviour of one type of visitor. And that type — the human with a browser — is no longer the only one reading your site. He’s no longer even the most important one.
II. Two kinds of visitors with different needs
Your site has two audiences today.
The first is the human. He arrives via a link, scans a page, forms an impression, clicks through or bounces. He cares about design, rhythm, imagery, a clear call to action. Your analytics is built around him. Sessions, scroll depth, time on page, bounce rate.
The second is a machine. A crawler from an AI company, a RAG system retrieving your page during a user query, an agent gathering information on someone’s behalf. That visitor doesn’t read. He extracts. He looks for structure, for facts, for relationships between concepts. He doesn’t leave the page — he takes it with him into an answer shown somewhere else entirely.
That second visitor is invisible in your dashboard. No session, no event. And yet he’s the one deciding whether your brand shows up in the AI answer being shown tomorrow to thousands of potential customers.
The shift is subtle but fundamental. Your website is being visited less. And read more.
III. Your website as reference document
In a world where people visited ten websites to make a decision, your site was one voice in a choir. Now, where a single AI answer summarises those ten websites, your site has become something else.
It’s a reference document. Not for the visitor — for the model reading on his behalf.
What that model picks up from your site determines what it says about you. If your product page lacks a clear explanation of what the product does, who it’s for, and how it differs from alternatives, the model has nothing to hold on to. It still answers the user’s question — but without you, or with a generic description in which you’re no longer recognisable.
A model uncertain about who you are won’t mention you. Or worse: it improvises. Hallucinates a fact that isn’t true. Confuses you with a competitor. The user doesn’t know the answer is wrong. You find out only when someone reaches out with an expectation that makes no sense.
This is the opposite of what marketers assume. The assumption: the website matters less when fewer people visit. The reality: that model is far stricter about clarity than the human who’s no longer there. What still feels like “enough context” for a scanning visitor is too vague for a model to do anything with. What you publish carries more weight, precisely because there’s less noise around it.
IV. The paradox of cutting back
Here’s the bitter conclusion. A website nobody visits feels like a cost that can be cut. But a website you scale back takes one thing down with it: the only place where you still have full control over what’s said about you.
Everything outside your site gets filtered, reworked, mixed with other sources. Reviews on external platforms, mentions in news articles, posts on forums, data on comparison sites — your grip on those is limited. You can respond, correct, influence. But you can’t write them.
Your own website is the only source you fully control. The only place where your version of the truth stands unfiltered. If you shrink it or neglect it, you don’t just lose traffic. You lose your position in the information stream AI models process. Your voice gets quieter, while everyone else’s voice about you keeps going.
So the real role of the website shifts. Not a marketing channel. An anchor point.
V. What this concretely demands
What does “a website read by models” mean in practice? Four things, in order of payoff.
Answers instead of pitches: Models pick up sources that answer questions. Not sources that pitch products. Every product page deserves a rewrite that starts from the user’s question, not from the brand’s message. A page explaining the difference between a fixed and variable formula, who each choice makes sense for, and what the trade-offs are — that’s usable material for a model. A page promising you’ll be done in a few clicks isn’t.
Structured data: Code via schema.org that makes explicit what your content means. Not just what it says, but what it is: a product, an FAQ, a price, an author. AI models don’t use schema directly as input, but search engines and parsing systems do — and through that detour it shapes which version of your content reaches a model. It doesn’t solve understanding. It helps with recognition: who you are, what you belong to, what kind of thing you’re describing. That reduces confusion and increases the odds the model builds a consistent picture of you.
A clear entity definition: Models don’t think in pages, they think in entities — clearly defined concepts with properties and relationships. Who are you, what do you belong to, what exactly do you do, where do you differ from alternatives? If that’s scattered across forty pages in thirty variations, the model can’t extract a coherent picture from it.
Consistent external signals: What others write about you carries more weight than what you write about yourself. Your own site is your foundation, but a foundation without walls around it collapses. Make sure mentions in independent sources — trade media, comparison sites, reviews — tell the same story you publish yourself.
A website doing these four things scores twice. It works for the human who does still arrive, and it reads well for the model reading in his place. A website that mostly pitches scores neither.
The question to ask of every page is no longer “does this sell?” but “does this answer a real question?”
VI. The metric your dashboard doesn't show
And then the question that sits beneath all of this. How do you know it’s working?
Sessions drop. Pageviews drop. Time on page drops. Everything your dashboard measures will look worse than it used to — even when your website is gaining relevance. That’s the perverse part of the shift. Your dashboard can tell you you’re losing at the exact moment you start winning.
The metrics that actually say something about what’s happening don’t live in Google Analytics. They live in AI answers that get logged nowhere. They’re not called session or conversion. They’re called mention frequency, citation quality, position in the answer, stability across models.
These aren’t alternatives to your existing metrics — they’re additions. Anyone looking only at sessions misses half. Anyone looking only at AI presence misses the other half.
The question to ask yourself by the end of this year isn’t: how many people visited my website? It’s: in how many answers about my sector did my brand appear, and with what weight?
That’s the new benchmark. And the website that makes it possible is the one you didn’t tear down.
