AI 101 — What is AI slop?
AI slop is AI-generated content that nobody asked for and nobody checked — text, images, audio and video poured out in bulk to win attention or ad money rather than to inform anyone.
Why the word is unavoidable now
For a while, AI slop was just internet slang. Then the dictionaries caught up. In December 2025 Merriam-Webster named slop its 2025 Word of the Year, defining it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence" — per Merriam-Webster's announcement, which listed the year's flood as "absurd videos, off-kilter advertising images, cheesy propaganda, fake news that looks pretty real, junky AI-written books, 'workslop' reports that waste coworkers' time."
A month later, the American Dialect Society voted the same word its 2025 Word of the Year, announced on January 9, 2026, noting that "AI slop" had merely been a nominee in 2024 — by 2025 the word could stand alone, the AI part understood. Kelly Elizabeth Wright, who presided over that vote, described a term that had travelled "from the pig sty, to the algorithm," and now breeds compounds: sloppunk, slopification, friend slop.
Where the term actually came from
The word is centuries older than the problem. Merriam-Webster traces it to "soft mud" in the 1700s, then food waste fed to pigs in the 1800s, then rubbish more generally. The AI sense was proposed in May 2024 by the programmer and writer Simon Willison, in a short post titled "Slop is the new name for unwanted AI-generated content". Willison's line was careful and still holds: not all AI-generated content is slop, and not all slop is promotional — but content that is mindlessly generated and pushed at someone who did not ask for it has a name.
The mental model: this is an economics story, not a genre
Slop is not a style. It is what happens when the cost of making content falls to roughly nothing while the cost of checking it stays human. Before generative models, every published page had passed through a person who decided it was worth making; that person was the filter, and their time was the bottleneck. Remove the bottleneck and one pipeline can produce thousands of pages a day, each one individually cheap enough that being wrong costs nothing — there is no byline to embarrass, no subscriber to lose, no editor to answer to.
That leaves three questions that separate slop from ordinary AI-assisted work: did a human read this before it was published? does it answer something anyone actually asked? and does it exist to inform, or to be counted, ranked and watched?
An analogy: the buffet, not the kitchen
A restaurant kitchen cooks each plate to order, and a chef answers for every dish that leaves it. A buffet refills its trays by the gallon, and nobody tastes each scoop. Both feed people. Only one of them stands behind every plate.
Slop is the buffet tray, scaled by machines and inspected by no one. The older meaning of the word — the pig slop that was whatever happened to be left over, poured into a trough in bulk — gives you the rest of the picture, with one change: the trough is now the recommendation feed, and the measurement is how long anything holds a thumb still.
How much of the web is it?
More than most people guess. The growth-research firm Graphite sampled 55,400 English-language articles from Common Crawl, the public web archive, and classified each one with three commercial AI detectors. Articles that were mostly AI-written climbed from a rounding error before ChatGPT to 35.9% within twelve months of its release, and have held at roughly half of everything published since early 2025 — 49.9% in the first quarter of 2026, according to Graphite's study.
The important caveat is in the same paper. Graphite's researchers found in separate work that AI-generated articles largely do not surface in Google results or in ChatGPT's answers, which suggests the place you meet slop is the endless scroll, the ad slot and the inbox — not the search box.
Common misconceptions
"AI slop means anything made with AI." No. Slop is defined by absent review, not by the tool. A translated contract a human checked, a first draft a reporter rewrote, an assistant's code suggestion a developer tested and merged: all AI-assisted, none of it slop. Willison, who uses language models daily, treats "don't publish slop" as a personal baseline, not an argument against the tools.
"You can tell by looking." Sometimes. The American Dialect Society also listed, among the year's most useful phrases, the flat statement "that's AI" — the reflex of doubting something that looks real. Note where that puts the verification work: on the reader, one image at a time.
"Better filters will fix it." Filters shift it. Slop is cheap to regenerate and adapts to whatever passes; the defences that hold up are provenance — labels and watermarks that try to mark machine-made content — plus one human who reads the thing before it goes out.
"It's only junk videos and fake news." The category reached professional life fast. In September 2026, mathematicians objected to what they called "slop mathematics" coming out of an AI-assisted research hackathon, and OpenAI pulled its sponsorship — we covered what happened next. Research slop is the same economics with a worse consequence: work that looks like a contribution and takes a specialist's week to disprove.
Where to learn more
Related reading: What is a deepfake? on the synthetic media that fills the same feeds · What is AI watermarking? on the labels meant to tell you a machine made it · OpenAI drops Caltech math hackathon after mathematicians cry 'slop' on slop arriving in science.
Where do you meet slop first — your feed, your inbox, or your search results? Write to us via our contact page.
