What’s left of a Website?

You can get an article like this one, maybe a little more background checked and neutral, just by asking ChatGPT. Or, better yet, you can ask ChatGPT to summarize this article.

So why read this? And if you don’t read this but use an LLM to summarize, then why should I write it nicely out in the first place. I could just put bullet points up here and it would be easier for both me and the LLM. That’s the shift that we are going through. And I’ve been writing about The Shift for a long time now, but always only bits and pieces. In this writeup I’ll try to put those together to draw a bigger, personal picture of the future of the web in an LLM focussed world. This is not about zero touch computer interaction or the web for communication or the internet in general – I have my separate thoughts on those topics. This is about how websites as we know them might evolve.

Simple websites

People have started noticing a few years back that most sites don’t need fancy layouts and that unnecessary design is often decreasing conversion. I am going to cite a paragraph from an old article of mine –

[Simple sites] load faster. You find what you came for. You read. You click. You buy. You leave. That’s conversion.

As much as page builders try to sell you on unlimited design flexibility, having a more complex design just for the sake of it doesn’t solve anything.

A restaurant site with ten full-screen sliders, a cookie banner and giant popup doesn’t bring anything to the table if the user ends up searching the opening hours for 5 minutes. […]. Design should follow purpose.

That doesn’t mean that all websites should look plain or stripped down. The web thrives on personality. Color, playfulness, creativity – all good and needed. But when form hides function, people leave.

And this sentiment about simple sites is only carried on in an LLM-focussed world. When less people visit your site, it doesn’t matter if your site looks fancy or not, because a crawler sees the HTML output of a site. And those crawlers have an easier time analyzing your site if they don’t need to parse 10 nested divs to get the price of a sandwich. As a result, the AI preferably uses content of clean, simple sites preferably which results in more leads.

And more and more people are noticing. Dave Bloom just replaced their complex sites with a simple one. He is right about it not mattering too much. His personal site isn’t selling anything, there’s no need to impress.

A current workaround/proposed solution for AI to “read” complex websites is to introduce /llms.txt files to all websites that explain the websites content for LLMs to understand. This is not only a bad solution because it separates content that AI sees from what we see, but also because it is a solution for a problem that shouldn’t exist in the first place.

The frontend-less web

If less and less people are accessing the websites themselves, eventually the frontend of websites might fade entirely. Why bother even building a simple frontend if people never visit your site? Why not just upload plain Markdown files directly?

That’s exactly where I see the future of web browsers headed. When a site is visited that doesn’t provide a frontend, it is the browsers responsibility to hold an on device AI to parse the provided contents and generate a frontend for the site. A simple one. Just something a person accessing a domain directly can read and see.

Single source of information?

I like the idea of everybody being able to set up their own web space. It provides freedom and means that nobody is in control. In an effort to find the most effective way of handling text and data in another writeup about the future of text in general, I wrote

We need a global knowledge cloud where the entirety of information is stored and where every LLM has access to. Think Wikipedia but bigger, and with private pages nobody can update. Nobody can own this knowledge cloud, just like nobody manages crypto. Updated in real time and all AI systems can access public entries – without needing to waste resources sifting through the entire web.

Nobody would own it, everyone’s in control, and yet it’s secure. But is all of that really necessary? Doesn’t Perplexity already search the web pretty well? Do we even want that giant brain? And could we even build this form of a knowledge cloud? Where is it stored? But then again, where is the BTC blockchain stored? I simply don’t know, but this is where we get into Web3 territory, a thing that went forgotten in the past years.

In the meantime, I like our domain based system.

Why publish at all?

There’s a million reasons to publish, but they all come down to two things.

One – getting heard, making noise. Two – making money.

And the internet is a great place for both of those things. As we’re moving away from classic backend-frontend websites towards a future where content is generated by LLMs, we need to make sure both of those things are protected.

Your voice

It started with the book press and continued with the internet. The fact that people can write something once and have it distributed to many is incredible. That I can write something potentially accessible to over 5.5 billion people is amazing, and, well, it needs to be protected. Websites are uncensored. They run on servers you can own, and nobody can tell you to take something down.

In an AI-first web, where end users no longer access individual servers directly and servers are instead used as information storage for AI responses (at least until we figure out a decentralized solution), people interested in my opinion no longer read my words – they see only the AI’s interpretation. And for that AI interpretation, there’s a thin line between manipulating and fact-checking that we need to figure out.

Making money

Ads won’t work in the future. And unless you offer exclusive, high-demand content, paywalls won’t either, same with selling data. None of these work if people never actually visit your domain. And if a site has a paywall, well, the browser will just move to the next website with the same information.

What we need is a Spotify-based licensing system where every site is paid per usage, very little, but with billions of people using AI that needs information, there might actually be a world where this works. Notice how Google AdSense currently doesn’t pay much per view and not much per click.

A small monthly fee, say EUR 2, for AI to use content from a particular source? Sure, that could work. Mostly, though, it would apply to exclusive content sites like Bloomberg with Apple leaks.

I’m positive on that one. It’ll further democratize publishing. A “this guy says this and that guy says that” model, where everyone publishes thoughts on their own server, just wasn’t possible before. And now it is.

Online stores?

One last thing – if the title made you wonder about online stores, that’s an interesting topic too. Sure, LLMs will be able to make purchases via shop MCP actions, but the real challenge is AI comparing prices across all providers in real time. There are and have been price comparison sites out there before – in Austria, we have idealo.at and geizhals.at – but most people don’t check. Or they don’t even know these exist. Or they already have an account at a provider company and don’t want to bother with creating a new one.

If everyone automatically gets the lowest price, it will create the biggest race to the bottom in prices so far. On-site upselling will be just as hard, same with user retention – unless we get AI to a state where you “open” a shop in the AI and then the shop site takes over the AI responses, but that would just be a worse user experience. So I don’t see that one as an option.