temu.swiss
AI slop is disruptively destructive for one's company.
Essay
6 Min. LesezeitVeröffentlicht 11. September 2026

Your Customers Have Already Stopped Reading

Thirty articles in ten seconds sounds like a productivity story. From the reader's side of the screen, it looks like something else entirely.

The moment my brain switched off

I remember the article but not the company, which is rather the point. I was a few sentences in when something in me tuned out. Not a decision. A reflex. My brain had recognised a pattern it had seen a thousand times that month: the tidy introduction, the confident tone, paragraphs that circled a topic without ever landing on it. It filed the piece as recycled and told me, quietly, that finishing it would give me nothing.

So I left the site. I trusted the company a little less, because they clearly had no idea what they had put in front of me. And a week later, the one thing I could still recall about them was that impression. Untrustworthy. Illiterate about the tools they were using.

That is the entire damage, and it happened in silence. No complaint, no line in the analytics that says "reader felt nothing." Just a person who was briefly a potential customer, and then was not.

Why nobody inside the building notices

The people who decide what gets published are, in most companies, not writers. They see what the machine returns and it looks fine. The grammar is correct, there is a heading and a conclusion, and it says something about their industry. What they cannot see, because seeing it takes a trained ear or at least an honest one, is that the text contains nothing a customer would actually want.

Then comes the extrapolation. If one acceptable article takes ten seconds, why not thirty? Type "write thirty articles about topics A, B and C," press Enter, and go for lunch. I understand the appeal. I have watched intelligent people do exactly this and feel like they had just discovered the printing press.

What they had discovered was a way to publish thirty pieces of nothing under their own name, at a speed no human could match.

The watch and the vocabulary

Think of any Swiss watchmaker with a real name behind it. The price is almost beside the point. What they sell is legacy: the watch a father gives to his son, which the son one day passes on again. Every word on their site and in their catalogues is chosen to carry that weight. Precise. Unhurried. You could recognise the voice with the logo covered.

Now feed that company a stream of generated articles. "Watch enthusiasts today are increasingly looking for timepieces that combine tradition with modern craftsmanship." The vocabulary that took decades to refine is replaced, overnight, by the average of everything ever written about watches on the internet. And the average sounds like nobody.

My intuition, and I will call it that because I cannot prove it, is that the more a brand rests on precision and heritage, the faster this kind of content hollows it out. A company selling plastic screwdrivers can probably survive a bland blog. A company whose whole promise is care cannot keep publishing evidence of carelessness.

Free content revealed something uncomfortable

A copywriter whose talk I watched recently put her finger on it. For years, companies explained their weak marketing as a resource problem: not enough time, not enough people to write. Then AI made writing free, and instead of a golden age we got a flood of sameness. The resource excuse vanished, and what stood exposed underneath was a skill problem. The tools guess from averages, which is why the output is so average. The people who never knew why one sentence worked better than another did not lose a skill to the machine. They never had it, and now everyone can tell.

Her filter for anything before it goes out is the one I would hand to every executive:

If AI could have written this, why is it going out under your name?

How I actually use it

I should be clear about something, because it would be easy to read this as a case against the technology. It is the opposite. I use AI for almost everything, including this article.

Here is how this piece came to be. The idea was mine; I decided the topic was worth the time. Then, instead of typing a prompt and accepting whatever came back, I asked the AI to interview me. It asked what I had seen, what I believed, where my examples came from, what I did not know. I answered at length. It built the body from those answers and shaped the language, and I went through the result and put my own thinking back in wherever it had thinned out.

The whole thing took an afternoon instead of four days. It sounds like me because the substance is mine. The machine did the part it is good at, which is speed and structure, and left the part it cannot do, which is having something to say.

Chess players have a name for this arrangement. After Kasparov lost to Deep Blue in 1997, he helped create a format where a human and a computer play as one team. They call it a centaur. The human brings judgement and instinct, the machine brings calculation, and the pair tends to beat either one playing alone. Human first, machine second. That order is the whole method.

One thing I would never do is publish a text the AI wrote without first reading it against my own point of view. If my thoughts are not in it, my name should not be on it either.

An honest gap

At Limen I spend my days looking at how AI search engines read and cite websites, so people ask me whether generated slop gets picked up by ChatGPT or Perplexity. I do not know. My guess is that the frontier models are good enough to recognise recycled content, but a guess is what it is, and I would rather say so than write a confident paragraph on the subject. Time and data will settle that question. The question of whether a human reader notices is already settled.

A test for tomorrow morning

You do not need a copywriter to run this. Take the last article your company published. Read it slowly, start to finish. Then ask yourself one question: did anything happen? A thought, or a disagreement, or a small flicker of feeling.

If nothing happened in your own mind, nothing happened in your customer's either. The piece was correct, and it was empty.

Stopping is the wrong fix. Feed the machine something worth amplifying instead: your opinions, your particular way of looking at what you do. Let it interview you. Let it draft and tidy and speed things up. Then read what it made and put yourself back in.

Feel free to use AI. It is a wonderful thing. Just make sure that when it speaks for your company, it is your voice it has been given, and not the internet's.

[@portabletext/react] Unknown block type "divider", specify a component for it in the `components.types` prop

This article reflects the author's own observations and opinions. It was drafted with the help of an AI system, in the way it describes.