Open any social feed, search engine, or marketplace today and you will find it: articles that say nothing in nine hundred words, product photos of hands with seven fingers, “news” videos read by synthetic anchors, albums by artists who do not exist. We have a name for all of it now, and the name won. Merriam-Webster made 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.” Australia’s Macquarie Dictionary picked “AI slop” for the same honor the same year. Dictionaries record complaints, they do not invent them. Two of them recording this one in the same year tells you the flood is real.
So here is a field guide to the problem: what slop is, why the internet keeps making more of it, what it costs, and what platforms, creators, and readers can realistically do about it.
What Exactly Is AI Slop?
The term did not come from a lab or a press release. It was popularized in 2024 by the poet-technologist known as deepfates, who predicted it would do for unwanted AI content what “spam” did for unwanted email. Developer Simon Willison cemented the definition a month later with a test that is still the best one we have: slop is AI-generated content that is unrequested and unreviewed.
“Not all promotional content is spam, and not all AI-generated content is slop. But if it’s mindlessly generated and thrust upon someone who didn’t ask for it, slop is the perfect term for it.” — Simon Willison, May 2024
That framing matters, because the lazy version of the term (“anything made with AI”) is useless and unfair. A programmer who reviews every line an assistant suggests is not producing slop. A journalist who uses transcription tools is not producing slop. Joel Lewenstein, head of product design at Anthropic, describes the same smell from inside the industry: “generic content, undifferentiated content, something that optimizes for volume and passive consumption over everything else.”
So the test is not was a model involved? It is was a human paying attention? Slop is what happens when the answer is no, at scale.
Why the Internet Is Drowning in It
Slop is not an accident of bad models. It is the rational output of three forces stacked on top of each other.
1. Marginal cost collapsed to zero. Before generative AI, producing a thousand articles, thumbnails, or tracks required a thousand human-hours. Now it requires a script and an API key. Anything whose production cost approaches zero gets produced until it is worthless. That is not cynicism, that is economics.
2. The attention economy pays for volume, not truth. Recommendation algorithms optimize for engagement, and engagement does not check provenance. Content farms, SEO operations, and even state-backed influence campaigns figured this out. Graphika researchers tracking propaganda networks found they had embraced AI not for sophistication but for throughput: unconvincing synthetic anchors, clunky translations, fake news sites that accidentally publish their own prompts in headlines. “It might be low-quality content, but it’s very scalable on a mass scale,” as Graphika analyst Dina Sadek put it. Slop does not need to fool everyone. It needs to exist in enough quantity that some of it sticks.
3. Nobody is filtering at the pipe. The numbers show how fast the flood rose. Research firm Graphite estimated that AI-generated articles on the open web surpassed human-written ones in November 2024, barely two years after ChatGPT launched. Deepfake videos grew an estimated 550% between 2019 and 2024. By mid-2024, slop channels made up roughly 10% of YouTube’s fastest-growing channels despite policies against “inauthentic content.” Spotify has had to purge millions of bot-generated tracks. When generation is free and distribution is unfiltered, the default state of the internet becomes noise.
A fourth danger moves slower than the rest: model collapse. Today’s slop is tomorrow’s training data. Every generation of unreviewed machine output that enters the corpus makes the next generation of models slightly more fluent at reproducing nothing.
What Slop Costs
It is tempting to laugh at slop and move on. The AI animals and seven-fingered hands are genuinely funny. But the costs land on real people:
- Trust. Once anything can be faked cheaply, everything becomes slightly deniable. Real events get dismissed as AI; fake events enter the news cycle before anyone checks. NPR spent 2025 documenting AI videos that went from app to viral to presidential repost in days.
- Human creators. Slop crowds out the work of people who spent years getting good. Georgetown marketing professor Ziwei Cong expects the internet to bifurcate: free tiers drown in generated filler while human-made work retreats behind paywalls. That is not a hypothetical; it is already the business model of several publications.
- Workers. “Hiring slop” (AI-generated résumés fired at every opening) and “workslop” (AI-generated reports that waste coworkers’ reading time) turned slop from a consumer problem into an internal tax on organizations.
- AI itself. Lewenstein’s warning is the most self-interested and most honest one in this debate: if the public’s only contact with AI is slop, “that’s what they will associate with AI, and that will be a detriment and a disservice to all of the positive outcomes that AI can generate.”
How to Fight AI Slop
There is no single fix, and anyone selling one is selling something. What works is pressure at four layers at once.
1. Platforms: hit the incentives, then filter
The cheapest lever is money. As researcher Kate Conway summarized it: “One way to regulate AI slop is to make it harder to profit from. If it can’t be monetized, the incentive to produce it at such a relentless rate disappears.” Demonetize mass-produced inauthentic content and half the slop economy evaporates overnight.
The second lever is labeling plus provenance, and here the record is mixed:
- Pinterest added “AI modified” labels, built classifiers to catch content with no metadata markers, and went one step further than most: a user control to reduce generative AI pins in your feed.
- LinkedIn and Snap began cracking down on generic mass-produced content in 2026 without banning AI tools outright.
- Google’s SynthID Detector lets anyone upload media and check for Google’s invisible watermark; over 10 billion items have been watermarked since 2023.
- C2PA, the provenance standard founded by Adobe, ARM, BBC, Intel, Microsoft, and Truepic, attaches signed metadata at capture time so you can verify who made something and whether AI touched it.
- The EU AI Act will require digital watermarking for many categories of AI-generated content from December 2026, turning provenance from a courtesy into a legal obligation.
Two caveats, though. Detection tools are probabilistic, not oracles: they produce false positives, and Meta infamously stamped real photographers’ work “Made by AI” before quietly renaming the label. A binary “AI or human” gate ends up punishing humans. And labels without user controls are theater. As The Verge kept pointing out through 2026, almost no major platform lets you simply filter AI-labeled content out of your feed; DeviantArt’s “suppress AI” toggle is the rare exception. Labeling tells you what something is. Filtering lets you decide what you see. We should demand both.
2. Creators and developers: adopt a personal baseline
Willison’s baseline is the whole ethic in three words: don’t publish slop. In practice, if you ship words, code, images, or audio:
- Attach your name to what you publish and stake your credibility on it. Anonymity plus automation is the slop factory recipe.
- Use AI as a draft tool, never as a publisher. Review everything before it reaches another human. Unreviewed output is the definition of slop, regardless of how good the model is.
- Disclose meaningful AI involvement and attach provenance where you can (C2PA credentials, SynthID, or a plain honest sentence).
- Optimize for the one reader who needed this, not the thousand impressions that tolerate it. Cite sources. Show your work. Specificity is the fingerprint of attention, and attention is the one thing generators cannot fake at scale.
3. Readers: curate ruthlessly, verify cheaply
You cannot out-scroll the flood, so change what feeds you:
- Move to curated pipes. RSS feeds, newsletters, small personal sites, and communities with editors are structurally harder to flood than algorithmic feeds. The “small web” is now a defensive technology.
- Learn the tells, but treat them as hints, not verdicts. Wikipedia’s WikiProject AI Cleanup maintains a list of chatbot signatures: hedged promotional adjectives (“breathtaking”, “vibrant”), stacked conjunctions (“moreover”, “furthermore”), unsourced superlatives, and the infamous em dash overuse. Wikipedia itself warns these must never be the sole basis for removal, and neither should they be for your judgment. The stronger tell is structural: no named sources, no specific numbers, no opinion that could be wrong.
- Verify before sharing. Reverse image search, check the account’s history, look for provenance metadata. Sharing is distribution; you are part of the pipe now.
- Pay for human work. Subscriptions and purchases are the only signal the attention economy cannot spoof. If you want human-made content to survive outside paywalls, fund it directly.
4. Organizations: build an immune system, not a ban
Wikipedia is the case study. Faced with a wave of machine-written articles, its volunteer community responded with what Wikimedia’s Marshall Miller called an immune-system reaction: speedy deletion for incoherent unsourced pages, a cleanup project documenting AI tells, and tooling like Paste Check that asks contributors who paste walls of text whether they actually wrote it. Note what Wikipedia did not do: it did not ban AI, and it did not let stylistic tells alone decide guilt. Policy plus tooling plus human judgment, in that order.
Companies can copy the pattern: require disclosure and review for anything published under the company name, make “who verified this?” part of the definition of done, and treat AI output as input to a human process rather than as a deliverable. The goal is not zero AI. It is zero unreviewed AI leaving the building.
The Long View
Every media revolution bred rubbish before it bred art. The early web was flooded with junk pages; photography was dismissed as mechanical kitsch; synthesizers were going to kill music. Scientific American’s retrospective on slop makes exactly this point: the flood is real, but so is the pattern of each flood eventually producing new tools, new filters, and new art forms. The word “slop” itself is one of those tools. Before “spam” entered general use, it was not obvious to everyone that blasting strangers with unwanted mail was bad behavior; naming it made the behavior discussable, then filterable, then socially expensive. Slop is having the same career.
Instagram’s Adam Mosseri admitted that “authenticity is becoming a scarce resource.” Scarcity is where value moves. The internet of the next decade will probably have two layers: a free, flooded, mostly synthetic surface, and a smaller verified layer where named humans stand behind what they made. Which layer you live in is, to a surprising degree, still a choice: what you filter, what you fund, what you review before you publish, and what you refuse to scroll past without checking.
The pipe is not going to stop pouring. You still get to choose what you drink from it. Be picky. Fund humans. Don’t publish slop.
Sources
- Ars Technica — Merriam-Webster crowns “slop” word of the year as AI content floods internet
- ABC News — ‘AI slop’ crowned word of the year 2025 by Macquarie Dictionary
- Simon Willison — Slop is the new name for unwanted AI-generated content
- Wikipedia — AI slop
- The Conversation — What is AI slop? A technologist explains this new and largely unwelcome form of online content
- MarketWatch — AI slop is taking over the internet. And it’s here to stay
- NBC News — Online propaganda campaigns are using ‘AI slop’, researchers say
- NPR — 2025 has seen an explosion of AI-generated slop
- TechCrunch — Pinterest launches new tools to fight AI slop
- NBC News — ‘AI slop’ crackdown: LinkedIn, Snap target low-quality gen-AI content
- The Verge — Let us filter AI slop, you cowards
- The Verge — If Big Tech cared about fighting AI slop, we wouldn’t be drowning in it
- TechCrunch — Google’s new SynthID Detector can help spot AI slop
- Fortune — Businesses are declaring war on AI slop. They are fighting a losing battle
- The Verge — How Wikipedia is fighting AI slop content
- Scientific American — AI Slop: How Every Media Revolution Breeds Rubbish and Art