--- title: "What does \"AI slop\" mean, and is it the same thing as machine voice?" url: https://deadwritingtheory.com/blog/what-does-ai-slop-mean/ summary: "AI slop is low-quality content made in bulk with AI and pushed at people who never asked for it. It is a verdict on quality. It does not prove who wrote a page, or that the web is converging." published: 2026-10-27 updated: 2026-10-27 author: "Jack Stovell" publisher: "Adapt Progress Evolve Limited" language: en-GB --- # What does "AI slop" mean, and is it the same thing as machine voice? *By Jack Stovell. Published 27 October 2026.* AI slop is low-quality content made in bulk with AI and pushed at people who never asked for it. It is a verdict on quality. It does not prove who wrote a page, or that the web is converging. Short answer: the word does three jobs at once. It says the content is poor, that it was made in bulk, and that a machine made it. Each of those needs different evidence, and the cases on this site can speak to only two of them, with caveats. ## The short answer: what slop means The tidiest definition is a dictionary's. Merriam-Webster chose "slop" as its 2025 Word of the Year and defines it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence." The looser web usage adds intent. On 8 May 2024 Simon Willison wrote up a post by @deepfates which predicted that "slop" would become "the term for unwanted AI generated content", the way "spam" became the word for unwanted email. Willison added the scope that matters: "Not all promotional content is spam, and not all AI-generated content is slop." His test is whether it is "mindlessly generated and thrust upon someone who didn't ask for it." Put the two together and you get a working definition: low-quality material made cheaply and in volume with AI, and pushed at people who did not ask for it. ## Where the word came from and who uses it I will not give a coinage date. Willison's post is the earliest primary source I read, and he was quoting someone else. What can be checked is the speed: by 2025 a dictionary had made a social media coinage its word of the year. The meaning is less settled than the dictionary makes it look. A literature review on the MINT Lab site, "AI Slop: Definitions and Normative Status" (credited to "Minty", 4 March 2026, and listed among the lab's agent reports), finds that "formal academic definitions remain sparse". It also notes that the split between Willison's framing, which is about the practice of pushing unreviewed output at people, and the dictionary's, which is about the content, "has not been resolved". So treat slop as a working term with a shared core. It is not a technical standard. ## Three claims hiding in one word When someone calls a page slop, they usually mean three things: 1. **Quality.** It is poor, padded or empty. 2. **Production.** It was made in bulk, cheaply, for traffic or volume. 3. **Origin.** A machine wrote it. They arrive together, so the word feels like one accusation. They are tested separately. Quality is a judgement, made by a reader, an editor or a panel. Production is about volume, cost and motive: how many pages, how fast, and who profits. Origin is provenance: did a model write this text? Any combination exists. Content farms turned out human-written junk at scale long before ChatGPT. A person can use a model carefully and publish something good. The word glues the three claims together, and that lets the origin claim ride in on the back of the quality complaint. ## Why "slop" is a verdict on quality, and quality is not what this site measures The [theory](/theory/) this site tests is narrower: has the web's writing collapsed into one machine voice? That splits into two questions the evidence keeps mixing up. Prevalence asks how much text is machine-made. Convergence asks whether writing is becoming more alike. Neither asks whether the writing is any good. Quality and convergence can even pull in opposite directions. In the experiment behind [Better stories, more alike](/cases/better-stories-more-alike/), Doshi and Hauser concluded: "With generative AI, writers are individually better off, but collectively a narrower scope of novel content is produced." By a quality test those stories are the opposite of slop. By a convergence test they show narrowing. Then there are the word-level tells. Plenty of people spot slop by vocabulary. Juzek and Ward, in "Why Does ChatGPT 'Delve' So Much?" (COLING 2025), looked for the cause of that overuse and report: "We fail to find evidence that lexical overrepresentation is caused by model architecture, algorithm choices, or training data." Their model testing is "consistent with RLHF playing a role", which is a lead and falls short of a settled answer. This site's own study found many of the same words were already common in human writing, press releases above all, before ChatGPT ([Where the AI-isms came from](/cases/where-did-ai-isms-come-from/)). A tell with a contested origin is weak evidence of who wrote a page. ## What the cases can say about slop, and what they cannot **Production and origin: something.** [The websites nobody wrote](/cases/the-websites-nobody-wrote/) counts sites dominated by machine text and checks whether they have a financial incentive, which is as close as the record gets to the "made for clicks" part. [Half the articles, and a plateau](/cases/half-the-articles-and-a-plateau/) estimates the share of new English articles that are primarily AI-generated. Neither is peer reviewed, the second is a vendor study, and both rely on AI detectors. Every detector-based estimate carries a false-positive caveat: some human pages get counted as machine pages. Even with perfect detectors, those numbers would measure prevalence. A count of machine-written pages is no count of slop. It says nothing about quality, or about whether anyone wanted the page. **Quality: nothing.** No case on the docket grades writing. A machine-written page can be accurate and useful, and a human one can be dire. **Convergence: a separate question.** A flood of slop would not prove that the web sounds the same, and a web that sounds the same would not have to be slop. Be careful with studies that claim to identify slop. Madler's "SlopShape: Identifying AI-Generated Commercial Web Content" (arXiv:2609.15369, September 2026) is a single-author preprint from Sitefire, a vendor. It compares generated mirrors, not what was actually published. That makes it a test of whether model rewrites can be told apart from human originals. Useful, but a different thing from measuring slop in the wild. When a study says it can spot slop, ask what it compared. ## How to use the word without overclaiming Keep saying it if you like. Just know which claim you are making. - If you mean the content is bad, say so, and be ready to defend it on quality. - If you mean it was mass-produced, point to the volume or the business behind it. - If you mean a model wrote it, you need provenance. A favourite word or a stray em dash is not provenance. - If you mean the web is turning into one voice, that is a convergence claim, and it needs its own evidence. For a lighter look at the word in the wild, see [the post on the joke site](/blog/your-ai-slop-bores-me-what-it-shows/). ## Sources - [Merriam-Webster, "2025 Word of the Year: Slop"](https://www.merriam-webster.com/wordplay/word-of-the-year) - [Simon Willison, "Slop is the new name for unwanted AI-generated content" (8 May 2024)](https://simonwillison.net/2024/May/8/slop/) - [MINT Lab, "AI Slop: Definitions and Normative Status" (credited to Minty, 4 March 2026)](https://mintresearch.org/reports/ai-slop/) - [Juzek and Ward, "Why Does ChatGPT 'Delve' So Much?", COLING 2025, arXiv:2412.11385](https://arxiv.org/abs/2412.11385) - [Madler, "SlopShape: Identifying AI-Generated Commercial Web Content", Sitefire, arXiv:2609.15369 (preprint)](https://arxiv.org/abs/2609.15369) ## Questions people ask ### What does AI slop mean? Merriam-Webster defines slop as "digital content of low quality that is produced usually in quantity by means of artificial intelligence." The looser web usage adds that nobody asked for it: Simon Willison's test is whether it is "mindlessly generated and thrust upon someone who didn't ask for it." ### Is AI slop the same as any AI-written text? No. As Willison put it, "not all AI-generated content is slop." The word adds a quality judgement and a mass-production claim to the origin claim. A careful, useful text written with a model is not slop in the usual sense. A human-written content farm page can be. ### Is "slop" an official or established term? It is established in the dictionary sense: Merriam-Webster chose it as its 2025 Word of the Year. It is not a technical standard. A March 2026 literature review on the MINT Lab site found that "formal academic definitions remain sparse" and that rival framings have not been reconciled. ### Can a detector tell you whether a page is slop? No. At best a detector estimates whether text was machine-made, which covers only the origin claim, and detector-based estimates carry a false-positive caveat. Whether the page is poor or unwanted is a judgement no detector makes. ### Does slop mean the web is converging on one voice? No. Slop is about quality and intent. Convergence is about whether writing is becoming more alike, and prevalence (how much is machine-made) is a third question. Each needs its own evidence.