The Site Without a Site: A New Way to Think About the Web

The other day I was looking for a vendor for a fairly specific service. A trivial thing, in theory: understand what a company does, who it works for, roughly how much it costs, whether it has experience in my sector. I opened the first Google result and ended up in a maze. Full-screen cookie wall, then a banner with a 20% discount if I subscribe to the newsletter, then a live chat opening on its own asking if I need help, then an autoplay video with epic music telling the company’s “vision” with words like “innovation”, “excellence”, “tailored 360-degree solutions”. I scrolled for a good three minutes trying to understand concretely what they do. I couldn’t.

The site was beautiful, mind you. Smooth animations, careful typography, professional photography. But by the end of the visit I knew less than before.

Then I did something I do more and more often. I took the URL of the site and fed it to Claude, asking: tell me who they are, what they do, who they work for, what their strengths are. The answer was bleak. Claude pulled out a handful of vague slogans, a few repeated keywords, no concrete cases, no numbers, no information that would let you form a judgement. The site was excellent at making a distracted human click, terrible at letting a machine understand what the hell that company does and whether it’s reliable.

And there I stopped to think. Because if I, a professional in the field, am doing this operation more and more often—if I use an LLM as a first filter to decide whether to dig further—how many other people are starting to do the same? And above all: how many decisions are already being made by automated systems reading your site instead of a human being?

The question that drives all this thinking is one, and I’ve been carrying it around for weeks: what if the next site shouldn’t be designed for the user looking at the screen anymore, but for the visitor who has no eyes?

I say “the visitor who has no eyes” and I realise it sounds like a provocation. But it isn’t, or at least not only. The first reader of your site today is a Google crawler. The second is an agent composing answers for someone who asked a question. The third is an LLM that has to summarise who you are, what you do, why it should trust you. Only after, maybe, comes the human. And that human, more and more often, arrives already with an opinion formed by what the machine told them. If the machine didn’t understand anything about you, the human won’t even get to see the site.

I’m not talking about SEO. SEO is part of the conversation, sure, but it’s a part we know by now and that has become almost a commodity. I’m talking about something different: machines that have to build a mental model of your company. Not a ranking, not a position in a SERP. A mental model. Understand who you are, what you actually do, who you do it for, with what results, with what limits. It’s an enormous difference. An SEO crawler looks for keywords and structure. An LLM looks for meaning, coherence, credibility. And if your site is a monument to emotional design but a desert of real content, that mental model will be empty. Or worse, it’ll be wrong.

I realise that to get here I have to first explain how the web “for humans” broke. Because it isn’t that one day someone decided to build incomprehensible sites. It happened gradually, decision after decision, layer after layer, until creating a monster no one consciously designed.

It started with cookie banners, and that was fine. Then came the newsletter pop-ups. Then push notifications. Then live chats. Then autoplay videos. Then mobile-app interstitials. Then discount banners. Every single element had its reason, its business case, its A/B test that proved a 3% increase in some metric. But the sum of all those 3% increases produced an experience that became, for many sites, genuinely hostile. Not only for machines. Also for humans, if we’re honest. But for machines particularly, because all that visual and interactive noise isn’t simply annoying—it’s opaque. An LLM can’t close a pop-up. It can’t scroll past a video. It can’t click “No thanks” on the newsletter. It sees the page’s source code, and that source code has become a battlefield where real content is buried under layers of JavaScript, tracking, dynamic components, lazily loaded content.

Then there’s the “conversion design” problem, and here I touch a raw nerve because I know that world well, I’ve worked in it. Every pixel optimised for micro-conversion. Every sentence designed to generate a click, not to explain something. Every page structured as a funnel pushing you toward an action, not as a document helping you understand. The result is that many corporate sites have become lead-capture machines but stopped being information sources. And when an LLM reads a page that’s essentially a long contact form with some slogans around it, what should it understand? That the company makes “innovative solutions”? That it’s “industry leader”? That it offers “tailored services”? They’re phrases that mean literally nothing. No reasonable human would take them seriously, let alone a model trained on billions of texts.

Then there’s the technical side, what I see from my vantage point as someone moving between code and strategy. Heavy single-page applications, with all content generated client-side via JavaScript. Texts appearing only after three interactions. Crucial information hidden in accordions no crawler expands. Duplicated landing pages with micro-variations for each campaign, creating insane semantic noise. Menus with twelve items pointing to pages with five paragraphs of pure filler each.

As a product owner, as a head of tech, as someone who every day mediates between business vision and technological reality, I see the paradox clearly. We built sites increasingly sophisticated from the perspective of human interaction, and in doing so we made them increasingly incomprehensible to the machines that are becoming our main intermediaries with the public. It’s like having built a beautiful shop but walled up the main entrance, forcing customers to go through a side alley full of obstacles.

And here I get to the concept that has been turning in my head for a while, and that I call, with some self-irony, the “site without a site”.

It isn’t a provocation for its own sake. It’s an alternative model of information architecture that, the more I think about it, the more it seems the only sensible one for what’s coming. The idea is simple in formulation: a coherent set of contents exposed to the web as if they were an API for humans and machines, with a minimal or even optional presentation layer. The keyword is “optional”. I’m not saying you don’t need a website. I’m saying the website should be the last layer, not the first. First comes information, structured, clear, complete. Then, on top, you put graphics, experience, brand.

It’s a total reversal compared to how sites get built today. Today you start from design, experience, “wow factor”. You write the texts after, often in a hurry, often delegated to someone who doesn’t know the company, often copied from competitors. Content is filler for a container designed before anyone knew what would go in it.

The “site without a site” starts from the other side. First you design the information graph. Who we are, what we do, for whom, with what results, with what processes, at what price, with what limits, with what evidence. You put it all down in simple, readable, textual format. You organise it so it makes sense even without a single line of CSS. Then, possibly, you build on top a presentation layer that makes everything more pleasant for a human to navigate. But that layer is a dress, not the body.

The parallel that comes to mind is technical documentation. Well-built open source projects have documentation that’s readable in plain text, in a terminal, in a README on GitHub. You can also make it prettier with a dedicated site, sure. But the substance is there before the dress. The “site without a site” is treating the entire corporate site as documentation. Not as a sales pitch, not as an interactive brochure, not as an emotional experience. As documentation. Clear, complete, navigable, comprehensible to anyone, human or machine.

I write these words and already feel the objections coming. But I’ll get to them in a moment, because first I want to push the metaphor all the way.

If the site is documentation, then content is data. Not data in the technical sense of database tables, but in the sense that every piece of information is an object with clear properties. A service has a name, a description, a target audience, a process, a price, limits, success cases. A case study has a client, a problem, a solution, measurable results. A team member has a role, skills, an experience. If you structure content this way, something interesting happens: it becomes reusable. The same content that feeds the website can feed a chatbot, a commercial PDF, a tender response, an LLM prompt, a marketplace listing. It’s native multichannel, not multichannel forced after the fact.

And the way of writing changes too. When you know your text will be read by a machine trying to build a mental model of your company, you stop writing slogans and start writing answers. Answers to precise questions: who is this service for? How does it work? How much does it cost? What does it include and what doesn’t it? What results have you achieved? What are your limits? That last question is the most powerful, and almost no one includes it on their site. Saying what you don’t do is incredibly precious information, both for a human evaluating whether to contact you and for an LLM deciding whether to recommend you.

How I’d actually build it

If I had to do it tomorrow for a client, and I’ve thought about it a lot, I’d start from three layers.

The first is a pure knowledge base. Content in markdown, organised by entity. A folder for the company with history, values, team. One for services, with a card for each one always structured the same way. One for case studies. One for processes. One for FAQs—the real ones, not the marketing-invented ones. One for policies. All in plain text, readable without a browser, comprehensible without visual context.

The second layer is the semantic structure. Coherent taxonomies, explicit relationships between entities, URL naming that makes sense, microcopy designed to answer questions and not to impress. This is the layer that lets both a human and a machine navigate the information and understand how the pieces connect.

The third layer is the exposure. A minimal site, even static, that adds no frills but makes navigable what already exists as a knowledge base. From this same layer you can generate feeds, APIs, structured files for agents. The website becomes one of the possible interfaces, not the only and not even the main one.

I know what you’re thinking: this works for a tech company, not for a fashion brand, not for a restaurant, not for a company that lives on image. And you’re right, in part. But only in part. Because the fashion brand also needs an LLM to know how to describe who it is, what it does, what its positioning is. The restaurant also needs that when someone asks an assistant “where do I eat well in Milan for fish?” the system finds clear and reliable information. The image-driven company also needs its image to be comprehensible beyond the visual. The presentation layer will be different, richer, more curated. But the underlying information layer has the same needs.

There’s an aspect of all this I find counter-intuitive and therefore particularly interesting: designing for those who have no eyes also improves the experience for those who do.

Think about it. If you force yourself to write content an LLM can understand, you’re writing clear, concrete, structured content. No empty slogans, no roundabout phrases, no fluff. An LLM doesn’t fall in love with your graphics, but rewards coherent text with examples, real cases, numbers, limits. Know who else appreciates that? People who have little time and want to quickly understand if you’re the right vendor. That is, basically all your potential clients.

If you design clean architecture for machines, you’re eliminating useless clicks, redundant pages, contorted paths. You’re creating what in SEO is called canonical structure, but applied to the entire site narrative. And you know what? It’s the same thing a human user would want: find what they’re looking for without getting lost in a labyrinth.

If you remove useless decoration to make the site readable to a crawler, you’re also making it faster, more accessible, more usable on mobile, more usable for people on slow connections or older devices. Accessibility and machine-readability are almost the same thing, seen from different angles.

And yet there’s a real tension I don’t want to hide, because it would be intellectually dishonest.

Every time you add a pop-up, you maybe improve newsletter signup by 2%, but you make the page opaque to every crawling, parsing, sequential-reading system. Every time you write “clever” but empty copy—great for the internal pitch—you make it impossible for a model to understand what you actually do. And if the model doesn’t understand, it won’t cite you, won’t propose you, won’t recommend you. Every time you build an “experiential” design that breaks the mould, with content fragmented and scattered across JavaScript components loading at different moments, for the machine that becomes background noise, a puzzle to reconstruct without the box image.

I’m not saying any decision made for the human is wrong. I’m saying there’s a trade-off almost no one is calculating. And that trade-off, with the passing months, leans more and more toward the machines. Not because machines are more important than humans, but because they’re becoming more and more, and better and better, the channel through which humans get to you.

At this point I already hear the objections. I hear them because I made them to myself, and because I know the design and marketing world well enough to know where they’d hit.

“This kills the brand.” No. The brand isn’t the gradient, isn’t the micro-animation, isn’t the parallax scrolling. The brand is the coherence of the message, the quality of promises kept, the clarity with which you communicate who you are. In fact, I’ll say more: a site full of visual fireworks but with empty content damages the brand, because it creates expectations the real experience doesn’t meet. The strongest brand is the one that says true things, clearly, and keeps them.

“It all becomes cold and technical.” This makes me smile, because it’s exactly the opposite. When you remove slogans and stock phrases, what’s left is space for an honest narrative. Real cases with names. Stories of projects with their problems and their solutions. Concrete details showing real competence, not declared. The “warmest” sites I’ve seen weren’t the ones with stock photos of people smiling in offices. They were the ones where you understood exactly with whom you’d work and how.

“But the competition has super-cool sites.” And that’s exactly the point. If everyone has super-cool sites and everyone says the same things with the same words, how do you differentiate? Maybe the best way to stand out is to have a site that doesn’t give human visitors a headache and at the same time becomes an ally of the models that have to talk about you. While the competition invests in animations, you invest in content. While they optimise for the wow, you optimise for understanding. It’s a bet, sure. But it’s a bet that seems less and less risky.

If you take this vision on board, and I know it’s a big if, here’s what changes in practice.

UX stops being a manipulative funnel and becomes a clear, predictable path. Every element that doesn’t add real information gets eliminated. Not because it’s ugly, but because it’s noise. Pages don’t get duplicated for every campaign, content doesn’t get fragmented into ten micro-sections with catchy little headings. You build canonical, linear paths that lead the visitor, human or machine, from question to answer in the most direct way possible.

SEO completely changes skin. You move from keyword stuffing to entity and relationship modelling. Content has to make sense even read in plain text, without style, without visual context. It has to answer real questions, not queries invented by a tool. And FAQs stop being that section at the bottom of the page no one reads and become the informational core of the site, because they’re exactly the format an LLM processes best.

Content follows strict internal guidelines. Every service has to explicitly answer: what it is, who it’s for, how it works, what it includes, what it excludes, how much it costs, what evidence there is that it works. If you can’t answer all these questions clearly, maybe the problem isn’t the site. It’s that you haven’t thought hard enough about what you offer.

Metrics change. Less obsession with pageviews and time on page, which are vanity metrics in most cases. More attention to the quality of incoming requests. Are the leads aligned with what you actually do? Do clients arrive already knowing what to ask for? Is there coherence between what the site promises and what the client expects? Those are the metrics that count.

I want to tell a story that, even simplified, is very close to situations I’ve seen with my own eyes. A B2B services company, fifteen people, with a site redone two years ago by a good agency. Flashy site, huge fonts, corporate storytelling, hero video with drone, “our values” section with hand-drawn icons. All beautiful. But if you asked anyone, including the founder, to explain in three sentences what the company does by reading only the site, they couldn’t. The service pages were full of words like “integrated approach”, “end-to-end solutions”, “strategic partner”. Real information: zero.

They ran an experiment. They took all the site text, put it in a file, gave it to an LLM asking: who are we? What do we do? Who are we the right choice for? The result shocked them. The model had understood they did “consulting” of some kind, maybe in digital, maybe in communication, it wasn’t clear. It hadn’t identified any specific sector, any distinctive competence, any concrete case. In practice, the €80,000 site told the same story as ten thousand other identical sites.

They redesigned everything starting from content. They rewrote every page answering the questions: what we do exactly, for whom, how, how much it costs, what we don’t do. They documented case studies with real numbers: problem, solution, measurable result. They put in a list of sectors where they have experience and, equally important, those where they don’t. The design became minimal, almost spartan. The copy became direct, at times brutal in its honesty.

The result wasn’t only a better ranking and better tool comprehension. Different leads started arriving. More qualified, more informed, who already knew what to ask. Sales calls became shorter because clients arrived with a precise idea of what to expect. The conversion rate rose not because the site was more persuasive, but because it attracted the right people and turned away the wrong ones. Which is exactly what a good site should do.

The site as persistent prompt

There’s an aspect of all this that pushes me to look a little further ahead, and that strikes me as the most strategic of all.

Your site is becoming a persistent prompt. It’s the text that trains LLMs on how to talk about your company. Every word you publish, every piece of information you make available, every absence you leave, contributes to forming the representation language models have of you. If you don’t say it clearly on your site, someone else will make it up. Or worse, the model itself will make it up, filling the gaps with what it finds elsewhere, or with the statistically most probable solution, which is almost never the right one.

Think about how many interactions are already happening without the user visiting your site. Someone asks an AI assistant: “which company do you recommend for this kind of service?” The assistant consults its sources, including your site, and formulates an answer. If your site is a monument to emotional storytelling but contains no concrete information, the answer will be vague, generic, or will exclude you entirely in favour of a competitor with a less beautiful but clearer site.

More and more often the flow will be: client talks to agent, agent reads the site, agent synthesises an answer, client decides based on that synthesis. Your site won’t be the client’s destination anymore. It’ll be a data source for an intermediary. And like any data source, it’ll be judged on the quality, completeness, and reliability of the information, not on the beauty of container and contents.

This makes the clarity of raw content even more strategic than any campaign, hero video, polished storytelling. Your raw content is what the world will see of you through the machines. It’s your face for those who have no eyes.

If you’ve made it this far, I propose an exercise that’s also a slightly merciless call to action. Do it today, it takes ten minutes.

Take all the text of your site. Just the text, without graphics, without layout, without colours. Put it in a file and give it to an LLM. Ask it: who are we? What do we do? Who are we the right choice for? Why should a potential client choose us?

If you don’t recognise yourself in the answer, the problem is serious. It isn’t an SEO problem, isn’t a design problem, isn’t a marketing problem. It’s an identity problem. Your site doesn’t know how to tell who you are, and in a world where LLMs are increasingly the first point of contact, that means no one will know who you are.

The question to ask isn’t “how do we make the site more engaging?” anymore. The question is: how do we make sure anyone, human or machine, understands at a glance who we are and why we matter?

If your most important visitor has no eyes, maybe it’s time to stop designing only for those who look and start designing for those who understand.