A big part of my job is looking at how websites are put together. Not the visual design, the structure underneath. How sections connect, how a visitor moves from one piece of content to another, how the whole thing holds together as an information system.
The recurring thing I notice is not bad design. It is that most sites barely have a structure at all. Pages exist as a flat list pretending to be a hierarchy. Sections that should connect, don’t. User paths lead into dead ends. The navigation is something nobody really thought through.
And then I remember something uncomfortable. A website is the visible layer. The part the company actually shows to the outside world. The part they presumably cared about the most. If even this is a mess, what do you think the data inside the company looks like?
Most companies would rather not look there.
What I keep seeing in the visible layer
Here is the recurring pattern from the audits. A site with a few hundred pages and no real topology. The blog talks about one set of topics, the product pages talk about another, the resources section is a third island, and there are almost no real connections between them. From any given page, the path forward is unclear. From the homepage, the path inward is unclear. The whole thing is a collection, not a structure.
When I bring this up, the conversation almost never turns toward fixing it. It turns toward generating more. Can we use AI to write fifty more articles. Can we use AI to make the existing pages more “discoverable to ChatGPT and Perplexity”. Can we add an AI search box to the site.
I get why those are the questions. They are the visible, project-shaped things you can put on a slide. Fixing the underlying topology is slow, unsexy, and shows up nowhere in the quarterly review.
But you cannot fix navigation by adding more rooms to a house that has no hallways. And the AI search engines that everyone is suddenly trying to please prefer the same thing humans prefer: clear structure, real connections, and content that knows what it is about. The cleanup work and the AI-readiness work are the same work. People just don’t recognize it because one of them sounds boring.
If this is what I see at the visible layer, on the part of the company that was designed to be navigated, I have to assume the internal data is in worse shape. And the internal data is what most AI projects are actually supposed to use. In a way, this is the organizational version of the context management I wrote about in episode 5 (and the file structure in episode 8). Same idea, just at the company level instead of the project level.
The reverse-order problem
There is a reason the data ends up like this, and it is not laziness. It is the order things happened in.
Most companies did not decide what AI was for and then go buy it. They bought AI first, and now they are looking for a problem to point it at.
I understand how it happened. The pressure to “do something with AI” was enormous and very fast. Boards asked for it, investors asked for it, competitors put out press releases, and nobody wanted to be the laggard. So the order got flipped. First, sign the contract. Then assemble a task force to figure out what we are actually doing.
The result is predictable. When you start from the tool instead of the problem, you end up trying to make the tool fit somewhere it doesn’t belong. You also never get around to the unglamorous question underneath: what data do we have, and what shape is it in. That is the same question the website avoids by generating more pages. Skipped at the page level, skipped again at the boardroom level.
The result is the same too. The moment the first project disappoints, the conclusion becomes “AI doesn’t work for us”, when the honest version would be “we never figured out what we wanted it to do”. What quietly damages companies for years is not the failed AI project itself. It is the disappointment that follows, which makes leadership skeptical of the next, better-designed attempt.
The alliance of blindness
If leadership bought the tool without a clear picture, you might expect at least the people below them to be the only ones in the dark. They are not.
Most employees I talk to are at least a little anxious about AI at work. They have read the headlines, they have used ChatGPT, and they are not sure where the line is between “useful tool” and “thing that makes my role redundant”. That is a real feeling and I am not going to pretend it isn’t.
What is harder to see from the inside is that the people who bought the AI often have no clearer picture than the employees do. They signed a contract under pressure to move fast, they integrated something hoping it would help, and now they are also waiting to see what it actually does for them. The tools are genuinely useful. The hard part is figuring out where they fit.
So you get this strange situation, half comforting and half worrying. Comforting, because the boss is also fumbling. Nobody is sitting on a secret playbook. Worrying, because a leader who does not understand the limits of a tool will set unrealistic expectations for the people using it. “Just use the AI, it should be fast” becomes the new version of “make it pop”.
The shared confusion is real, and pretending it isn’t makes everything harder.
The hype is also AI now
And when people are confused, they go looking for clarity. The trouble is where they go looking. This is the part where the whole thing eats its own tail.
Imagine you are an executive trying to make sense of all this. You want to do the responsible thing. You open LinkedIn to see what serious people are saying. What you find is a feed of confident posts about AI tools, written in a very particular voice, illustrated with very particular images, citing very particular case studies. And if you look closely, a lot of that content is itself generated by AI, to sell other AI products, to people who are trying to research AI.
I am not saying this to be cynical. It is just the situation. The signal-to-noise ratio in the place where most decision-makers get their information is noisy right now, partly because the same technology they are trying to evaluate is also being used to write about it. Trying to think clearly inside that loop is hard.
I notice it when I talk to clients. Half the questions they bring are not really their questions. They are questions a LinkedIn post planted in their head a week ago.
What I would actually suggest
I am going to keep this short, because the post already gestures at the answer.
If you are a person who works with AI tools at your job: become a deliberate user before you become a fast user. There is no rush to adopt every new app and every new prompt template. This is the “productive procrastination” I wrote about in episode 1. The people who will get the most out of these tools over the next five years are the ones who took a few months to actually understand what the tools are doing, not the ones who installed the most plugins in the first quarter.
If you are running a company or a team: put the boring work first. Know what data you have, know where it lives, know what shape it is in, fix the obvious gaps. Then talk about what AI you want to layer on top of it. The order matters more than the budget.
None of this is going to make a great slide. That is part of why so few people are doing it.
The phrase “your data is not ready for AI” has caught on lately. More and more people are saying it, and that is a good thing. The part that is still rare is doing something about it: the slow, unglamorous cleanup that nobody can put on a slide.
Time is on the side of the people who build slowly. Not the loudest, not the fastest, not the ones with the most polished landing page. The ones who took the unsexy step of cleaning the foundation before adding another floor. That part has not changed, and I don’t think AI is going to change it either.