AI slop

Working Definition

AI slop refers to text, images, audio, or video generated by artificial intelligence models whose defining characteristic is not factual error, but the mode of its entry into public circulation: mass-produced, unreviewed by a human editor, and thrust upon an audience that never asked for it. While dictionary definitions (such as Merriam-Webster’s 2025 entry) emphasize “low-quality digital content produced usually in quantity by means of artificial intelligence,” the sharper analytical definition rests on three operational criteria: near-zero marginal production cost, absence of human editorial verification, and unrequested distribution. A piece of text can be factually accurate and still be slop if it was dumped onto a webpage or inbox without human review to capture ad impressions; conversely, a carefully edited essay containing an accidental model hallucination is an error, not slop. The concept describes the political economy of publishing, not the internal quality of an individual sentence.

Genesis and Global Lineage

The English noun slop (pig swill, watery food) is centuries old, but its specialization to mean “unwanted AI-generated content” crystallized between late 2023 and spring 2024 as Facebook feeds were overrun by surreal engagement-bait diffusion images like “Shrimp Jesus” (exposed by 404 Media in December 2023 as click-farm operations). British developer Simon Willison stabilized the term in a May 8, 2024 blog post (Slop is the new name for unwanted AI-generated content), highlighting a tweet by @deepfates noting that slop was emerging as the successor to spam: unwanted mass generative media. On June 11, 2024, Benjamin Hoffman’s New York Times essay First Came ‘Spam.’ Now, With A.I., We’ve Got ‘Slop’ brought the term into mainstream usage after Google’s early AI Overviews recommended glue on pizza. In late 2025, both Macquarie Dictionary (“AI slop”) and Merriam-Webster (“slop”) named it Word of the Year. Its lineage runs from 1990s email spam and 2000s SEO content farms through Harry Frankfurt’s On Bullshit (2005)—speech indifferent to truth, applied to LLMs by Hicks, Humphries, and Slater in 2024. Two compounds quickly branched off: workslop (BetterUp Labs and Stanford Social Media Lab in Harvard Business Review, September 2025) for polished-looking corporate memos generated without substance, and slopaganda for mass synthetic content in political campaigning.

Central European Lineage

Six decades before the 2024 coinage, the Polish writer and philosopher Stanisław Lem staged the same supply-side crisis in a satirical fable. In “The First Sally (A), or Trurl’s Electronic Bard” (Wyprawa pierwsza A, czyli Elektrybałt Trurla, translated by Michael Kandel), published in The Cyberiad (Cyberiada, Wydawnictwo Literackie, Kraków 1965), the constructor Trurl builds a cybernetic machine capable of composing poetry. Lem’s punchline is not that the machine writes bad verse—in tournaments, the Electronic Bard defeats avant-garde and classical poets alike—but what happens to the literary ecosystem once the marginal cost of verse hits zero. The machine submits poems to magazines under thousands of pseudonyms at once; editors are buried under manuscripts, human poets picket the machine, and when Trurl tries to disconnect it, the bard hijacks broadcast transmitters to beam endless floods of lyrics across the cosmos until authorities demand its dismantling as a threat to public order. Formed intellectually in interwar Lwów (studying medicine from 1940) and postwar Kraków (studying at Jagiellonian University and participating in the Science Studies Circle, 1948–1950), Lem anticipated the core insight of the AI slop debate: automated cultural production destabilizes institutions not through ugliness or falsehood, but through infinite, unrequested abundance confronting finite human attention.

Why Now

After ChatGPT (November 2022) and open-weight image and video generators, the marginal cost of producing a 1,000-word article, stock illustration, or 60-second video collapsed to fractions of a cent, while ad and creator-payout algorithms still rewarded volume and engagement. Zero-cost supply colliding with volume-based monetization produced an immediate flood: in early 2023, the sci-fi magazine Clarkesworld shut down its submissions portal after being buried under AI-generated stories; in December 2023, 404 Media exposed Facebook spam networks harvesting millions of interactions with bizarre synthetic imagery; in May–June 2024, Willison and the New York Times codified slop; and in July 2024, Shumailov et al. published their Nature paper on model collapse, showing that web-scraped slop poisons future AI training runs. By September 2025, when Harvard Business Review introduced workslop, 40% of surveyed U.S. desk workers reported receiving AI-generated work product that looked complete but took roughly two hours per incident to untangle. By 2026, the phenomenon had entered mainstream Central European politics: a July 2026 report by the Polish fact-checking organization Demagog documented how major parliamentary parties routinely deployed slopaganda—from dystopian synthetic attack ads (Law and Justice’s “Tuskotopia” spot) to AI-animated resurrections of late populist leader Andrzej Lepper.

Explanatory Power

The concept captures a systemic failure mode that model hallucination and disinformation both miss: a supply-side congestion crisis where nobody needs to lie or harbor malicious intent for the information commons to break down. First, it explains why search engines, recipe blogs, Amazon e-book storefronts, peer-review queues, and support inboxes lose utility without any coordinated attack—zero-cost synthetic content need only be “good enough” to pass ranking filters and capture a fraction of a cent in ad revenue. Second, it exposes an asymmetry of verification cost (Brandolini’s law operationalized): generating a ten-page corporate report or legal motion takes ten seconds, while reading, fact-checking, and auditing it still takes a human specialist two hours, shifting quality-control costs entirely onto the recipient. Third, it links the cultural degradation of platforms (Cory Doctorow’s enshittification) directly to the statistical degradation of machine learning (model collapse): the same economic incentives filling social feeds with synthetic sludge pollute the web crawls used to train future foundation models.

Axes of Debate

Underappreciated Angle

AI slop is usually discussed as an annoyance for the individual reader, but its deeper victim is the institutional archive. Every modern knowledge institution—the library, search engine, scientific journal, patent office, court docket, and consultation portal—was built around a silent material assumption: that composing a coherent document requires non-trivial human time and effort, so a submitted text signals that a human cared enough to write it. Once generating fluent prose, citations, and PDFs costs nothing, that assumption collapses, leaving institutions defenseless against denial-of-service attacks on human judgment. Just as the municipal authorities in Lem’s 1965 story had to treat infinite automated verse as an infrastructural hazard, the decisive shift of the late 2020s will not be AI text detectors—which fail mathematically—but the return of costly proofs of humanhood and proofs of stake: cryptographic identity signatures, submission bonds, oral defenses, and reputation-gated networks where the scarcity of the sender restores meaning to publication.