AI Slop Is Eating the Internet

Illustration: a mountain of steaming gray mush burying a smartphone, tiny arms of the phone waving for help, flat vector, warm reds/oranges

There's a word the internet settled on for the machine-generated gunk clogging your feeds, and it's AI slop. Not "AI content." Not "synthetic media." Slop. The kind of word you'd use for pig feed and cafeteria gravy. It arrived the way all good slang arrives — because nothing else fit — and within months it was on Wired covers, in Senate hearings, and on the lips of every product manager quietly asking their content team to "scale output with AI."

Here's the thesis of this entire publication in one sentence: AI is just spicy autocomplete. And spicy autocomplete, left unsupervised, produces slop.

What Even Counts as AI Slop?

AI slop isn't all AI-generated content. It's a specific failure mode: text, images, and video that gesture at meaning without carrying any. It's the LinkedIn post that says "In today's fast-paced world" and then says nothing for 400 words. It's the Pinterest pin of a rustic farmhouse with eight chimneys. It's the YouTube video titled "She Asked For DIRECTIONS, What Happened Next SHOCKED Everyone" narrated by a voice that has never once inhaled.

The tell is always the same: slop optimizes for the metric, not the meaning. Clicks, not insight. Impressions, not impressions that matter. It's SEO content from 2009, except now it writes itself in milliseconds and costs a fraction of a cent per page.

The numbers are genuinely absurd. Researchers estimated that by late 2025, AI slop made up a double-digit share of new text on the open web, with certain corners — product reviews, recipe blogs, news aggregators — pushing past 50%. Amazon's Kindle store drowned in machine-written "books" that no human will ever finish reading. If the current funding frenzy looks familiar, that's because it rhymes with the great AI jobs panic — infinite supply meeting finite attention, with valuations doing the hallucinating. Facebook's own recommender began pushing uncanny AI images of shrimp Jesuses and soldier-crab weddings to millions of confused aunts. Nobody asked for this. Everyone engaged with it.

The Slop Supply Chain

Slop doesn't appear by magic. It's a supply chain, and it's worth knowing the players:

The generators. Open-weight and closed models alike get press-ganged into content farms. The barrier to entry is a laptop and a scraping script. A single operator can run fifty "niche news sites" before lunch.

The distributors. Search engines, social feeds, and ad networks — all engagement-hungry, all easily gamed by volume. Google's core updates keep trying to demote slop, and they keep partially working, in the same way bailouts keep partially working: the water level drops, then rises again.

The launderers. This is the funniest part. Much AI slop exists to be ingested as training data for future models, which then produce slightly more slop, which then trains the next generation. It's a human centipede of content. Researchers call this "model collapse." Its close cousin is AI hallucinations — the model filling gaps with confident fiction, now at industrial scale.

The buyers. Ad networks still pay by the impression. Nobody at the agency asks whether the eyeball belonged to a human who wanted to be there. If the dashboard is green, the invoice gets paid. This is the economic engine of slop, and it's the only part that matters: as long as someone pays for impressions, slop is a rational business.

Why Fighting AI Slop Is So Hard

Every anti-slop measure creates a new slop market. Watermarking? Congratulations, you've invented a startup that sells watermark removal. AI detectors? They false-positive on human writing — especially by non-native speakers — and miss the good slop, because the good slop is precisely the stuff that reads human. "Verify it was made by a human" quickly becomes a CAPTCHA for content, and we all remember how CAPTCHAs turned out: a decade of clicking crosswalks while bots walked right past.

The deeper problem is taste. Slop thrives because most readers can't reliably tell, and many don't care. The average reader of a "10 Best Air Fryers" listicle doesn't need truth; they need a page that loads, lists ten things, and has a Buy button. The demand side of slop is us, doomscrolling at 1 a.m., and no algorithm can fix that.

The slop isn't coming. The slop is here. It has a Shopify store and a Substack.

The Cleanup Has (Quietly) Begun

Here's the real insight, and it's less bleak than you'd expect: slop is starting to devalue itself. Publishers discovered that AI-written content gets AI-written competition in nanoseconds, so the margin collapsed to zero. "Scale your blog with AI" worked exactly once — for the first person who did it. Everyone after that was farming the same plot.

Meanwhile, the premium for human signal is rising. Newsletters with real voices are booming. Communities with strong norms — certain subreddits, Discords, group chats — police slop aggressively, not with detectors but with social shame, humanity's oldest and most effective moderation tool. It's the same instinct behind the backlash to vibe coding disasters: people can smell unreviewed machine output. Getting ratio'd by your own community remains undefeated.

And the platforms are slowly learning the lesson that spam taught them in the 2000s: unpriced garbage eventually poisons the well for the ad business itself. When advertisers started realizing their "brand-safe" campaigns were running next to shrimp Jesus, the money started asking questions. Money asking questions is the closest thing the internet has to a immune system.

So What Do We Do With It?

Stop pretending there's a technical fix. There's a market fix: make slop expensive and make provenance valuable. Some concrete things already happening:

• Pay for the humans. If you like a writer, pay them. The price gap between "free slop" and "paid human" is the entire editorial industry's business model now.
• Demand provenance. Signed, verifiable, accountable authorship is coming back — not because it's noble, but because advertisers will eventually refuse to buy impressions they can't attribute to a person.
• Keep the bar high in your own circles. Every community that enforces "no low-effort AI posts" is a data source that stays clean. That's a moat.

AI slop is the internet discovering what happens when creation becomes free but attention doesn't. The slop era won't end with a bang or a ban — it'll end with a whimper, when the economics stop working and everyone quietly admits the pig feed was never food. Until then: we document the chaos.