Slop: A Keyword's First Two Years, Traced Through Dated Examples

Compare four definitions of one word, written within about nineteen months of each other. On 8 May 2024 Simon Willison called slop the term for "unwanted AI generated content", and wrote that it is "mindlessly generated and thrust upon someone who didn't ask for it" [1]. On 25 November 2025 the Macquarie Dictionary named "AI slop" its word of the year, defining it as low-quality content created by generative AI, "often containing errors, and not requested by the user" [2]. On 15 December 2025 Merriam-Webster chose "slop" and defined it as "digital content of low quality that is produced usually in quantity by means of artificial intelligence" [3]. On 9 January 2026 the American Dialect Society voted for "slop" as "low-quality, high-quantity content, most typically produced by generative AI", and added that it works as a combining form for "anything of little value generated in mass quantities" [4].

Read down the list and two things drop out. The clause about the reader's wishes ("unwanted", "didn't ask for it", "not requested") is gone by the third entry. And the machine, required in the first, becomes "most typically" in the last. My thesis is that these two losses are connected, and that they matter. Slop spread because it names something spam could not: content that readers don't want but platforms do. As the "unwanted" clause fades, the word is drifting toward "anything low-effort", and a word that covers everything cheap stops being able to accuse anyone in particular.

This is interpretation, not a finding. I read dated definitions and documents. I did not count usage, and I flag that limit again below.

What the word carried before the machine

The word is old and its baggage is physical. Dictionary.com traces it from a Middle English "mud hole" to swill fed to farm animals, then, from the 1950s, figuratively to "rubbish", and in the 1990s to "inferior media products" [5]. It reports that the AI sense emerged around 2022, with early uses on 4chan and in YouTube comments [5]. Wikipedia's account agrees on 2022 and adds Hacker News [6]. None of this is a coinage story. Willison's May 2024 post credits the Twitter account @deepfates with noticing the term emerging, and he has said elsewhere that it was in use before him [1][7].

What the farm sense contributes is a scene. Slop is what is poured into a trough. The eater is not consulted, and the eater is not the customer. That is already a more exact picture of a feed than "spam" gives.

The gap spam could not fill

Willison pitched slop as the spam of the AI era: as "spam" became the word for unwanted email, "slop" would be the word for unwanted AI content [1]. It is a good pitch, and it explains why the word stuck as an unpretentious, derisive noun. But look at what spam presupposes. Spam is sent. Someone, the spammer, pushes it at you, and the mail provider is, at least in principle, on your side with a filter. The grammar of the word sets up two parties and a defender.

Now take the case that did most to make the word famous in 2024, the AI images on Facebook. Josh Goldstein and Renée DiResta studied 125 Facebook Pages that each posted at least 50 AI-generated images. They found the Feed recommending unlabeled AI images to users who neither followed the Pages nor knew the images were AI-made, and one such post reached roughly 40 million views and more than 1.9 million interactions in the third quarter of 2023 [8]. The researchers, writing for a general audience, called this spam and scams [9]. Fair enough: the operators were spammers in motive, sending users to ad-laden sites [9]. But the delivery mechanism was the platform's own recommender, which they suggest may have been "organically promoting these posts" [9].

Here the spam model strains. The defender has become the distributor. The reader never opted in, and the platform's incentives, which are measured in interactions, ran against the reader's preference. Wikipedia's summary of the economics adds the producer's side: creators in developing nations can earn well from high volume because advertising on Facebook and TikTok pays more per US viewer [6]. The money moves through the platform at both ends.

My claim is that slop is the word for this triangle. It carries three roles that spam leaves implicit: a producer paid by volume, a reader who did not ask, and a platform that gets engagement either way. The trough does the last job. Nobody calls it spam when the thing arrives in your ranked feed, bearing the platform's own recommendation.

I hold this at moderate confidence, perhaps 0.6. The weakest part is causal: I cannot show that people reached for "slop" because "spam" failed them, only that the word's meaning fits the gap and that its early definers wrote the reader's non-consent into it.

Two years of dating, two kinds of loss

The sequence at the top is the evidence for the drift. Take each loss in turn.

The reader's consent drops out. Macquarie, in November 2025, still keeps "not requested by the user" [2]. Merriam-Webster, three weeks later, does not. Its definition is about quality, quantity and a technology [3]. Greg Barlow, the dictionary's president, explained the choice by calling it "part of a transformative technology, AI" that people find "fascinating, annoying, and a little bit ridiculous", and the dictionary noted a tone "less fearful, more mocking" [3]. A mocking tone is a fine thing for a word to carry. But mockery attaches to a quality, and the original complaint was about a relationship.

The machine drops out. The American Dialect Society's "anything of little value generated in mass quantities" is the widest wording yet [4]. Its examples include "sloppunk", "slopification" and "friend slop" [4]. I cannot tell from the source how each is used, but the combining form is plainly not tied to a model.

Other coinages show the same slide. Wikipedia lists "workslop", "scholarslop" and "Microslop" [6]. Workslop is the clearest case, because it arrived with a definition: in the 22 September 2025 Harvard Business Review article by Kate Niederhoffer and colleagues, it is AI-generated work content that masquerades as good work but lacks the substance to advance a task [10]. That is still a version of the triangle: a producer (a colleague), a recipient who did not ask, and a system (the office's tools and incentives) that rewards sending. A blog post about the study put the matter in its title: "bad 'study', but an excellent word" [11]. I have only read that title, not the argument under it, so I cite it for the title alone. The word travelled further than the evidence did, at least on that post's account.

Then the fourth step, which I can date: on 26 July 2026 Ben Geudens published "The terrible rise of 'human slop'", defining it as "low-effort, repetitive, emotionally manipulative content produced by humans who imagine that their hostility toward machines constitutes an original personality" [12]. Here there is no machine, and nobody is thrust anything. The target is people who post indignation. I am not ruling on whether the description is fair. I am saying what the word is doing: it is now an insult for a style, and it can be thrown at the people who adopted it first.

What the blunting costs

A critical word earns its keep by being hard to dodge and easy to check. "Spam" works because you can say what makes a message spam: unsolicited, bulk. "Slop" at its best offers a similar test. Did anyone ask for this? Who is paid when it appears? Did a platform choose to show it to me?

Once slop means "low-effort", the test turns into a taste judgment, and taste judgments are what the accused can return. Geudens's essay is the demonstration: it takes the word and points it at the people who most use it [12]. That does not make him wrong about any individual post. It shows the word has lost its direction.

Max Read offers the best case for the wide version, and it deserves a serious answer. He proposes that slop is that which is "fully optimized" to its domain to the point of texturelessness, anything "designed to be as easy as possible to produce, sell, and consume", and that it is a symptom of "the feed", the binge-watch and the subscription service, so that generative AI may be "a product of a culture of slop" rather than the reverse [13].

The strongest objection

Read's argument is the strongest objection to my thesis, because it says the drift is not a loss but a correction. If slop was always a feature of platform economics, then tying it to AI was the mistake, and the broader sense is the one that finally names the structure. The human-made content farm and the film made to a streaming formula are the same product as the Facebook image. I find much of that persuasive. Content built to be cheap to make and cheap to consume existed before 2022. Dictionary.com's date for the figurative "inferior media products" sense, the 1990s, supports it [5].

Here is where I part from it. Read's definition is about the object: how optimized it is. Mine is about the relationship: who wanted it, and who profits from its arrival. The two come apart in tests. A uniformly competent, fully optimized pop single that you chose to play is slop by Read's definition, but nobody forced it on you. A lumpy, strange, individually made AI image that Facebook shows you unasked, to harvest your attention for an ad page, is not slop by his definition, but it is exactly what the word was coined for. The relational definition keeps the target. The object definition keeps the aesthetic complaint, and aesthetic complaints are cheap to make about anything.

The crux, then, is whether a word is more useful as a diagnosis of quality or as a charge about distribution. I think the second is scarcer. English has plenty of ways to say "bad". It has few for "bad, unrequested, and promoted for profit".

If I am right

If the relational reading is right, the word will keep widening in the ordinary way, and the places that still do the work of the original will start to need another term. Expect a qualifier to appear for the strict sense, as "AI slop" already does, or a new noun entirely. Dictionaries have, in effect, chosen the wide meaning by dropping the consent clause, which suggests the narrow meaning will survive mainly in the writing of people who study platforms. And since the narrow sense points at feeds that pay for volume, the criticism that depends on it will go on needing precise nouns: the recommender, the payout, the page.

What would change my mind is counts. If a dated sample of headlines and posts shows the "unwanted by readers" sense still dominating through 2026, my drift claim is too strong. If "slop" is applied to platform-boosted content and to ordinary human-made cheap content at similar rates, the relational sense has already thinned out. I have not run that sample. Test it on your own feed this week: take the next twenty things called slop and ask, for each, who asked for it, and who was paid when you saw it. How many survive both questions?