In the age of AI slop, judgement becomes the scarcest commodity

“Slop” is the swill you feed pigs. For a couple of years the word has meant mass-produced AI content: articles answering questions nobody asked, glossy images all alike, videos with synthetic voices and stock music, reviews written by no one. Kurzgesagt made a video with an eloquent title: AI slop is destroying the internet.

They’re right to worry. But I think the more interesting conclusion lies elsewhere.

The problem: an abundance of the plausible

The point isn’t that generated content is bad. It’s often correct, tidy, well written. The problem is that it’s average: the average of everything the model read. The most likely way of saying something, which by definition is the way everyone has already said it.

When producing text costs nothing, the web fills with average text. Finding an answer gets harder, not easier. The old “dead internet theory” (a web populated more by bots than people) sounded like a conspiracy fantasy; today it describes some corners of social media rather well.

And people notice. Emphatic openings, triplets of adjectives, conclusions that recap everything: we’ve learned to spot them, and when we do, we stop trusting.

The consequence: what’s scarce becomes valuable

Whenever something becomes abundant, value moves to what stays scarce. With AI slop, three things become scarce.

1. Judgement. Choosing what’s worth saying, what to cut, which of ten data points matters. AI generates options endlessly; deciding remains human work, and becomes the most valuable part.

2. Verification. In a sea of plausible text, verified text is a different product. A memo signed by a professional who answers for it, a figure with source and date, a review from someone who actually tried the product. Trust becomes the product.

3. Voice. A recognisable way of thinking and writing, with real experiences inside. No model can invent your case from yesterday, your mistake from three years ago, your unpopular opinion.

The After Skool channel puts it optimistically: the abundance of slop could spark a new “human renaissance”, because authenticity will become the distinguishing signal. I’m not sure about a renaissance, but the direction seems right.

This site is written with the help of AI, and I don’t hide it. Precisely for that reason I have rules:

Every number has a source and a date. If I can’t verify it, I don’t write it or I say it’s unverified.
Opinions are signed and labelled. Like this one.
Real examples. The notebook with 299 videos, the real cost of an invoice automation, the attribution error I found while checking: things that happened, not invented for the occasion.
Rereading and cutting. No text goes out as the model wrote it. Usually half the work is removing.

For anyone using AI to communicate (professionals, companies, creators) the advice is the same: use it for the mechanical part (structure, first draft, proofreading, translation) and add what nobody else can. If your text could have been written by anyone with the same prompt, you’re not communicating: you’re adding noise.

And labels?

The AI Act requires disclosure of deepfakes and of generated text published to inform the public, and from 2 December 2026 asks providers to mark content in a machine-readable way. Some models already embed invisible watermarks in text. Useful, but it won’t solve the problem alone: watermarks can be circumvented, and content can be true even if generated or false even if handwritten.

In the end, the real label is just one: who puts their name to it.

In one line

When everyone can generate everything, what counts is who chooses, who checks and who signs.

Sources

Informational content, current as of the date shown. Prices, models and rules change often: always check official sources. Written with the support of AI tools and reviewed by the author.