AI Slop: The Word That Named the Content Glut

On December 15, 2025, Merriam-Webster announced "slop" as its word of the year for 2025. Their definition: "digital content of low quality that is produced usually in quantity by means of artificial intelligence". The word was chosen because lookup volume on their site climbed throughout 2025, and what people were looking up was exactly what had happened: absurd videos, off-kilter advertising images, fake news that looked convincing, and digital books not worth reading.

What makes this word matter is not that it is new. "Slop" has been in English since the 1700s meaning soft mud, shifted in the 1800s to mean food waste fed to pigs, and then became a synonym for something worthless. What is new is that this shift happened again, and this time not into meat but onto screens.

There are two reasons this word is worth studying. First, it supplies language for something that previously could only be explained at length. Second, the dictionary's own choice contains a judgement: out of every word that rose in 2025, they picked the most dismissive. That is a signal about how the public evaluates AI output, and anyone producing content should take it seriously.

An origin with nothing to do with AI

The most interesting thing about this word is that it is not a recent invention. "Slop" was chosen precisely because the meaning is three centuries old, and the new sense attaches to the old one.

The path is simple and traceable. 1700s: thin mud, excavation leftovers. 1800s: food waste fed to livestock, a word referring to what animals will consume without any choice. Two centuries later: rubbish, or anything of no value. And 2025: cheap digital content arriving in very large numbers from machines.

There is a recurring pattern here. The word refers to something that was meant to be discarded but instead got consumed. And the right word for AI content was not "trash" or "garbage" but the word referring to animals with no mechanism to refuse. That the metaphor involves livestock is not incidental: it reads as a critique of the publishing industry itself.

Then there is the tone. "Slop" is mocking, not alarming. Compare it with the language usually used for AI threats in public discourse, which tends toward the dramatic. "Slop" sounds like people saying "yeah, we all know that is not real". That is exactly what you need to describe something that is extremely common and not worth reading.

What actually changed

The most common misunderstanding is that AI slop is just spam with a new name. Spam has existed since 1994, search engines have understood it for nearly two decades, and anti-spam handbooks are mature. What changed is the economics.

Before generative language models were available, producing content at volume required real time and real money. Hiring editors to write thousands of articles, photographing thousands of product shots, shipping thousands of units. That cost limited volume. Enforcement had an economic rationale, and penalties had clear economic limits.

When that cost fell to near zero, volume rose by orders of magnitude. Producing ten thousand pages takes hours and a small token bill, not months of work with editors and cameras. Once volume climbs far beyond what a search engine can evaluate, the threshold for enforcement changes.

The numbers above represent orders of magnitude, not a single paper's measurement. What matters is the pattern: the gap between the last two columns spans two orders of magnitude, and that is what changes behaviour.

This is why "slop" is technically not bad content but bad content arriving in volume disproportionate to the value that can be verified. One generic paragraph is a mistake. A hundred thousand generic paragraphs across different topics is a strategy. The word "quantity" in Merriam-Webster's definition is not incidental: it is the word that separates the two.

Google's rules, finally enforced

Google has had a policy against automatically generated content for years, and it was deliberately broad because it was hard to tell generic human-written content from machine-assisted content. What changed in March 2024 was that Google decided to measure something more specific: whether the content exists to manipulate rankings.

On March 5, 2024, Google announced a core update more complex than its usual core updates, plus three new spam policies at once. One of them is named scaled content abuse, defined as generating many pages whose primary purpose is manipulating search rankings rather than helping users.

Examples include using generative AI tools to produce many pages without adding value for users; transforming, translating, or obfuscating content from one source to create many URL variants; and creating many pages that make little sense to a reader but contain search keywords. Notably, the policy is deliberately vague about who produced the content. In the same announcement Google wrote that its long-standing spam policy is that the use of automation, including generative AI, is spam if the primary purpose is manipulating rankings. The operative phrase is "primary purpose", not "tool".

The official numbers are worth reading as a correction to a promise. The update was announced March 5, 2024, the rollout finished April 19, 2024, and on April 26, 2024 Google published an update: the actual result was 45% less low-quality content in search results, five percentage points better than the 40% they had projected.

One detail is rarely discussed: how it was presented. An analyst reading that announcement noted Google's blog post was worded remarkably carefully without ever naming AI. That was deliberate, for two reasons. Policies that name keywords are trivially evaded by swapping words, and acknowledging AI as the cause opens political room to argue about AI instead of about content behaviour.

Deeper dive: why volume is a technical problem

"Slop" is usually discussed as an ethics problem. What is rarely discussed is that very large volume destroys a resource: time and verification capacity.

A search engine does not merely crawl pages. It has to judge whether a page deserves to be shown, and that judgement needs references to check claims against. When one publisher generates ten thousand pages across ten thousand different topics, every page contains claims that need checking against something. Technically this turns the problem from "is this content true" into "do we have the capacity to check all of it".

The numbers above show shape, not absolute values. Production rises by a far larger factor than verification, and that gap is what makes selective verification the bottleneck.

For readers there is a more concrete cost. Reading generic content does not always feel like reading rubbish, it feels like reading something that answers nothing. And because AI output is so easy and so cheap, whoever writes the most produces the most volume without bearing the cost that used to make others stop. The content marketplace is flooded.

There is a simple calculation that often gets skipped. A human editor who needs six hours to write one researched article still loses to a team that can produce a hundred articles a month, not because those articles are more honest, but because nobody is checking them. Parallel production systems have no answer to the question of who stands behind any given claim. When no one is accountable, nothing can be held accountable, and that is the core of this problem.

Before that, a comparison worth making, because three strategies produce three different kinds of content and only one of them is sustainable:

The first and third rows use the same tools. The difference between them is not the AI, it is how much effort goes into checking what came out.

The uncomfortable edge: small site owners who write one article at a time with real research now compete against sites publishing thousands of unoriginal pages a day. That competition is not fair, and Google knows it. The scaled content abuse policy targets exactly this: reducing the flood, not banning AI from writing.

Deeper dive: separating slop from legitimate use

This last part matters because there is a risk you draw the wrong conclusion from this article. Slop does not mean using AI. That mistaken conclusion is worth correcting directly.

Simon Willison, who documented the term, gave a careful judgement himself: not all promotional content is spam, and not all AI-generated content is slop. What separates them is not whether AI was used, but whether the output adds value for the reader or only adds volume.

That difference is not technical and it is not about AI. It is about whether there is someone accountable for what gets published. In 2026, that is the scarcest and most valuable thing there is.

Slop will not disappear because platforms get better filters. It will disappear, if it does, when the cost of writing with care rises back up, when you have to stand behind every claim you make with your own name. Every careful claim needs someone who can be held responsible for it, and that is something no batch process can generate.