Pillar · The dependency lexicon

What is AI slop?

This is a reference page, not a daily dose. Read it slowly, then go do something without a screen.

Definition

Slop is the internet's own word — contemptuous, agricultural — for the rising tide of machine-made filler that arrived once generating a thousand words or a thousand images cost roughly nothing. It is not a judgement about AI as a tool. Careful work is often machine-assisted and nobody calls that slop. Slop is what happens when the content exists because it could be made, not because anyone needed it.

Three properties show up almost every time: it is produced at volume, it is produced without a person accountable for its accuracy, and it is optimised for a distribution channel (a search engine, a feed, a marketplace) rather than for a reader.

Where slop comes from

Nobody wakes up wanting to make slop. The incentives do the work. A programmatic ad pays the same whether a human or a model wrote the page. A marketplace ranks a listing the same whether the photo is real or rendered. A recruiter's filter counts a cover letter the same either way. Once generation is free and distribution pays by the unit, the rational move — for a content farm, a dropshipper, a stretched marketing team — is to make more.

Then it compounds. Models trained on the web start ingesting the slop the previous generation produced, and the average gets flatter. That feedback loop has a name in the research literature: model collapse. In practice, it looks like the whole internet slowly converging on the same beige paragraph.

What it looks like

• Search results that answer nothing. Eight hundred words restating the question, one sentence of substance, a recipe's worth of preamble before the number you came for.
• Shopping listings and reviews. Products photographed by no camera, five-star prose that names no feature, books credited to authors who never existed.
• Feed filler. Uncanny images engineered for reflex engagement, engagement-bait questions, "quotes" nobody said.
• Inbox and workplace slop. Expanded emails that the recipient will ask a model to summarise back down — a full round trip with no information added.
• Bug reports and submissions. Plausible-looking security reports, papers and pull requests that waste real reviewers' hours.

How to spot AI slop

No single tell is proof. A stack of them usually is.

• No accountable author. A stock-photo byline, an "editorial team", or a name with no other work anywhere.
• No verifiable sources. Confident numbers with no origin, or citations that loop back to other slop.
• Hedged, symmetrical prose. Everything is "a double-edged sword"; every section is exactly the same length; nothing commits to a claim you could disagree with.
• Impossible publishing rhythm. A site posting forty in-depth articles a day, on unrelated topics, in flawless nobody-voice.
• Images that fall apart up close. Text on signs, hands, jewellery, reflections and background crowds are still where generators break.
• No specificity you couldn't invent. Real expertise carries prices, dates, place names, failures. Slop stays general because generality never gets caught.

Why it matters

The harm of slop is not any individual bad page — it's the tax. Every real answer now sits behind ten plausible fakes, so the effort of finding anything trustworthy goes up for everyone, permanently. Search gets worse, marketplaces get worse, and the reflex to verify erodes precisely when it's most needed.

There's a second cost, quieter: when the flood is indistinguishable from the work, people stop making the work. Why research a thing carefully if it lands next to forty machine-made copies of itself?

And when slop acquires an agenda, it becomes slopaganda — the political strain of the same disease.

The hAIroin take

hAIroin is a satirical daily blog about our addiction to artificial intelligence. Slop is the drug in its cheapest, most cut form: we asked for infinite content, we got infinite content, and now we can't taste anything. The satire is the needle; this page is the leaflet in the box.

Browse the doses filed under the AI slop tag, or read the manifesto for the worldview behind it.

Everything else worth keeping — the reading list, detection tools and a free printable field guide — lives on the AI slop resources page.

Breaking the loop

One practical step, today: pick the three sites you actually trust on a subject you care about and bookmark them. For one week, start there instead of in a search box. You will notice, quickly, how much of what you used to read was filler that happened to rank.

To go further, the Center for Humane Technology's Take Control guide is a solid, non-commercial place to start taking your attention back.

Side effects may include: impatience with listicles, a sudden interest in bylines, and the ability to smell a generated paragraph at range.

Français →