AI slop: how machine junk is flooding the internet
Fake images, empty articles, bot traffic, and the dead-internet feeling. Why generative junk exploded, what it costs search and trust, and how to browse through it.
Somewhere around 2024, people started noticing the internet felt off. Facebook filled with surreal AI images — Jesus made of shrimp, impossible log cabins, weeping soldiers with six fingers — each earning tens of thousands of likes from accounts that may or may not contain people. Google searches for simple questions began returning pages that read like they were written by no one, because they were. Amazon carried AI-written travel guides for cities their "authors" had never seen, and obituary spam sites raced to publish machine-generated tributes to strangers minutes after their deaths, to farm the search traffic of the grieving.
The internet found a word for all of it: slop. AI-generated content produced at industrial volume, optimized for clicks and ad pennies, with no human on either end who cares whether it is true, good, or real. It is the pollution problem of the AI era, and it is changing how the web works.
Why is there suddenly so much of it?
Because the economics flipped. Junk content always existed — content farms paid humans small money to write filler a decade ago. But a human still cost something per article, which put a floor under the flood. Generative AI removed the floor. One person with a laptop can now publish hundreds of articles, images, or videos a day for approximately nothing, and platforms pay out on views. When producing garbage is free and attention pays, you get exactly what the incentive orders: infinite garbage.
The numbers tell the story. Within roughly a year of the big image tools launching, researchers estimated AI had produced billions of images. Web analysts began finding that a large share of newly created pages showed machine-generated fingerprints. Musicians noticed streaming platforms filling with AI tracks under fake artist names. None of this needed an audience that likes slop — it only needed recommendation algorithms that mistake engagement for value, and elderly relatives who click.
Is the "dead internet" real?
The dead internet theory — the old joke that most of the web is bots talking to bots — was a conspiracy meme years ago. It is now an uncomfortable trendline. Bots generate the post, bots like the post, bots reply to the post, and ad systems pay the operator as if humans were present. Security firms have reported for years that automated traffic approaches half of everything on the web, and generative AI gave those bots faces, prose styles, and personalities.
The web is not dead — you are reading a page a human wrote, and humans remain very much online. What is dying is the default assumption of personhood. For thirty years, seeing a photo, a review, a comment, or a stranger's story came with the reasonable prior that a person made it. That prior is gone, and it is not coming back.
What does slop actually cost us?
Search gets worse: the answer you need is increasingly buried under machine-written pages engineered to rank rather than inform. Trust gets worse: every real photo can now be dismissed as fake, and every fake defended as real — scholars call this the liar's dividend, and politicians have already learned to use it. Creators get squeezed: illustrators, stock photographers, and writers now compete with infinite free imitation, some of it trained on their own work. And the models themselves are threatened by it: researchers showed that AIs trained repeatedly on AI output degrade — a finding nicknamed model collapse — which means the labs, too, now need to tell human writing from slop. The polluters need clean water like everyone else.
How do you live on the sloppy internet?
Adjust your instincts the way a city dweller adjusts to traffic. Treat engagement-bait images and too-perfect stories as machine output until proven otherwise, especially anything designed to make you angry or weepy in one second flat. Favor sources where a nameable human stakes a reputation — bylines, faces, institutions with something to lose. Expect verification to move from the content to the source: since the pixels can no longer prove anything, who published them is the remaining signal, which is why provenance standards and watermarking are being built into cameras and models now.
And notice the strange flip at the center of it: for decades, "content" was scarce and attention was cheap. Machines made content infinite, which quietly made the scarce things human again — judgment, taste, reporting, firsthand experience, and the simple fact of being a real person who was actually there. Slop devalued everything machines can make. Everything they cannot just got more expensive.