If “the internet is cooked” had a corporate deck, 2025 would be slide one. The word slop—once basically “wet leftovers for animals”—became a serious label for something we all recognize: cheap, mass-produced AI content that looks like information but behaves like noise. Merriam-Webster didn’t just nod at the trend; it crowned “slop” Word of the Year and formalized the definition as AI-produced low-quality digital content made in quantity. slop cheap, mass-produced AI content that looks like information but behaves like noise. Here’s the uncomfortable part: AI slop isn’t just a “quality problem.” It’s a systems problem: systems problem incentives (traffic, ad money, affiliate payouts)
distribution (recommendation engines)
low marginal cost (content is basically free to generate)
weak provenance (hard to tell what’s real, or who made it) incentives (traffic, ad money, affiliate payouts) distribution (recommendation engines) low marginal cost (content is basically free to generate) weak provenance (hard to tell what’s real, or who made it) who So, this isn’t a rant. It’s a field report: scale, impact, and governance paths that don’t rely on wishful thinking. scale, impact, and governance paths 1) What Counts as “AI Slop” (And What Doesn’t) Let’s separate terms that get mixed together: Deepfakes: synthetic media meant to deceive (often identity/voice/video).
Hallucinations: model errors (it makes up citations or facts).
AI slop: the broader category—content churned out cheaply at scale, with low effort and low information value. Deepfakes: synthetic media meant to deceive (often identity/voice/video). Deepfakes deceive Hallucinations: model errors (it makes up citations or facts). Hallucinations AI slop: the broader category—content churned out cheaply at scale, with low effort and low information value. AI slop That’s why the Merriam-Webster definition matters. It anchors the term in two critical properties: produced usually in quantity
low quality produced usually in quantity produced usually in quantity low quality low quality Intent can be malicious or merely lazy. Slop doesn’t need a conspiracy; it only needs a CPM. 2) The Scale: Are We Actually Past the Tipping Point? A lot of “AI dominates the internet” claims are vibes dressed as statistics. The cleaner way to talk about scale is: in specific measurable slices of the web, AI output is now comparable to human output. in specific measurable slices of the web, AI output is now comparable to human output Graphite analyzed a dataset of online English-language articles over 2020–2025, and Axios visualized the result: by May 2025, human-written articles were about 52%, while AI-written articles were about 48%. May 2025 52% 48% That’s not “the whole internet” (and the methodology matters), but it’s still a cultural threshold: search results become easier to manipulate
content farms become economically viable
“average quality” gets pulled down by volume search results become easier to manipulate content farms become economically viable “average quality” gets pulled down by volume Also, even Wikipedia’s “AI slop” entry frames the term around mass production and monetization dynamics. 3) Why 2025 Made Slop Inevitable: The Incentive Stack Three forces made 2025 feel different: 3.1 The marginal cost of content went to ~zero. If you can generate 1,000 “SEO articles” in an afternoon, content becomes a commodity. And commodities flood markets. commodity 3.2 Distribution is automated. Recommendation algorithms don’t ask “is this meaningful?”   They ask, “Will someone pause for 1.2 seconds?” 3.3 Monetization doesn’t care who wrote it. Ad networks pay for impressions. Affiliate programmes pay conversions. That’s a slop subsidy. This is why slop is not only on social media. It’s also in: search spam
product reviews
workplace docs
low-effort “expert” explainers search spam product reviews workplace docs low-effort “expert” explainers 4) The Damage: It’s Not Just Annoying 4.1 Workslop: The workplace version of slop. BetterUp Labs + Stanford Social Media Lab describe “workslop” as AI-generated content that looks polished but lacks substance, creating an “invisible tax” where colleagues spend time cleaning it up. The key takeaway isn’t the exact dollar figure. It’s the dynamic: AI doesn’t always remove work; sometimes it moves work downstream to the next person. AI doesn’t always remove work; sometimes it moves work downstream to the next person. moves 4.2 Marketplace trust is getting sandblasted. There’s measurable evidence of likely AI-written reviews climbing on major fast-fashion/low-cost e-commerce platforms. Originality.ai reports likely AI reviews about Temu on Trustpilot, rising to 10.90% in 2025 (and up sharply from earlier years).  Cybernews reports similar figures for Temu and Shein, framing this as a surge in AI-generated reviews. Temu 10.90% When reviews become synthetic, “social proof” collapses. And platforms pay the cost in customer support, refunds, and reputation. 4.3 Elections: Deepfakes as The Sharpest Edge of Slop. Not all slop is deepfake, but deepfake is the slop category that gets people hurt. During Ireland’s presidential race, multiple outlets reported an AI-generated video falsely claiming candidate Catherine Connolly had quit, prompting platform removals and legal/political scrutiny. Even if the election outcome doesn’t flip, the trust damage compounds: “I saw it with my own eyes” becomes unreliable
Rebuttals are slower than virality
cynicism rises (“everything is fake anyway”) “I saw it with my own eyes” becomes unreliable Rebuttals are slower than virality cynicism rises (“everything is fake anyway”) 5) The Governance Reality: The 3-Layer Strategy That Might Work We don’t “solve” slop with one law or one detector. We reduce it by attacking three layers: provenance, distribution, and economics. provenance distribution economics 5.1 Provenance: Labels, Metadata, and Traceability China’s Cyberspace Administration of China (CAC) published the Measures for the Labeling of AI-Generated Synthetic Content, with an effective date of September 1, 2025. Measures for the Labeling of AI-Generated Synthetic Content September 1, 2025 The directional lesson isn’t “copy China.” It’s that labelling becomes part of the infrastructure: labelling becomes part of the infrastructure visible labels for humans (“AI-generated”)
machine-readable metadata for platforms and auditors
rules against stripping labels visible labels for humans (“AI-generated”) machine-readable metadata for platforms and auditors rules against stripping labels 5.2 Distribution: Platform friction beats perfect detection. Detection is hard—and getting harder. Research on “faithfulness hallucination” detection is improving (e.g., FaithLens proposes joint detection + explanation). But even great detectors won’t catch everything, and false positives create backlash. So, platforms will need friction: friction throttle low-reputation bulk posters
rate-limit new domains that behave like content farms
reduce reach for unverified media in high-risk categories (health, elections)
“Are you sure?” prompts before resharing unverified clips throttle low-reputation bulk posters rate-limit new domains that behave like content farms reduce reach for unverified media in high-risk categories (health, elections) “Are you sure?” prompts before resharing unverified clips Friction is less glamorous than AI vs. AI, but it works because it changes behavior. 5.3 Economics: Stop paying for junk. If ad networks and affiliate programmes keep paying for slop, the slop machine will keep running. A practical governance move: require provenance and publisher reputation for premium ads
penalise repeat slop farms (not just single pages)
increase the cost of automation (verification, KYC, rate caps) require provenance and publisher reputation for premium ads penalise repeat slop farms (not just single pages) increase the cost of automation (verification, KYC, rate caps) This is the “follow the money” layer, and it’s the most neglected. 6) The Weird Twist: Even “Words of the Year” Became a Signal You’ll see sloppy claims floating around, like “The Economist picked slop as word of the year.” It didn’t. The Economist’s official Word of the Year piece for 2025 chose TACO (“Trump Always Chickens Out”). TACO But the broader cultural signal is real: Merriam-Webster picked slop.
Oxford chose rage bait—a cousin phenomenon of engagement-driven outrage loops. Merriam-Webster picked slop. slop Oxford chose rage bait—a cousin phenomenon of engagement-driven outrage loops. rage bait Together, they map a coherent 2025 mood:   we’re drowning in synthetic content and engineered emotion. we’re drowning in synthetic content and engineered emotion. 7) A Practical Playbook for Teams (Not Governments) If you’re not a regulator, here’s what you can do inside a company/product: 7.1 For Content Teams make “human-reviewed” a product feature
measure information density (how much substance per token)
require sources (or ban claims without citations) make “human-reviewed” a product feature measure information density (how much substance per token) information density require sources (or ban claims without citations) 7.2 For Product + Platform Teams treat provenance as a first-class signal
add friction for reposting and bulk upload
build reputation systems that can’t be trivially gamed by bots treat provenance as a first-class signal add friction for reposting and bulk upload build reputation systems that can’t be trivially gamed by bots 7.3 For Engineering Teams Shipping AI Features default to “assist, don’t autoplay”
add “verification steps” in workflows (citations, checks, validation prompts)
log and audit generated content downstream default to “assist, don’t autoplay” add “verification steps” in workflows (citations, checks, validation prompts) log and audit generated content downstream The goal is not “no AI.” The goal is no cheap amplification of low-value output. no cheap amplification of low-value output. Final Take: Slop Is an Incentive Failure Wearing a Tech Costume AI slop exists because it’s profitable to produce and cheap to distribute. So, the fix is unsexy: provenance infrastructure
platform friction
economic disincentives
and literacy (people learning what “synthetic” looks like) provenance infrastructure platform friction economic disincentives and literacy (people learning what “synthetic” looks like) If we do nothing, the web doesn’t “end.” It just becomes a place where signal costs money and noise is free. signal costs money noise is free And that is how a civilization ends up eating content troughs for breakfast.

This story contains AI-generated text. The author has used AI either for research, to generate outlines, or write the text itself. 

AI Slop in 2025: The Year the Internet Started Eating Itself

About Author

Comments

TOPICS

THIS ARTICLE WAS FEATURED IN

Related Stories

A Developer's Guide to Merging AI with the Spring Ecosystem

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

The Noonification: White Man (11/26/2022)

The Noonification: The Metaverse is a Sh*tshow (11/2/2022)

100 Days of AI Day 1: From Newsletter to Podcast, Leveraging AI for Audio Transformation

A Developer's Guide to Merging AI with the Spring Ecosystem

The Noonification: How Often Do NFTs Pass The Howey Test? (1/13/2023)

Darwin's Hybrid Intelligence to Align AI & Human Goals for Startups & VCs

The Noonification: White Man (11/26/2022)

The Noonification: The Metaverse is a Sh*tshow (11/2/2022)

100 Days of AI Day 1: From Newsletter to Podcast, Leveraging AI for Audio Transformation

Light-Mode

Classic

Newspaper

Minty

Dark-Mode

Neon Noir

Minty

HN StartUps