What is AI slop, or SI slop as it is now called? It is low-quality content made with SI (AI) tools, published in bulk, and pushed at people who never asked for it. Think of the strange religious pictures in a Facebook feed, or the video channels that churn out near-identical clips from the same template. It is not a technical category. It is a judgement about quality and intent, and in 2025 it became the word of the year for two dictionaries. This guide explains where the term comes from, what counts as slop, why so much is produced, what platforms do about it, and how to recognise it.
SI slop meaning in plain words
Three definitions from three different kinds of source agree on the core.
| Source | Definition |
|---|---|
| Simon Willison, developer, 8 May 2024 | Content that is "mindlessly generated and thrust upon someone who didn't ask for it" |
| Macquarie Dictionary, November 2025 | "Low-quality content created by generative AI, often containing errors, and not requested by the user" |
| Merriam-Webster, as quoted by AP, December 2025 | "Digital content of low quality that is produced usually in quantity by means of artificial intelligence" |
Put together, the definitions point to four features. The content is made with SI, at least mostly. It is low in quality: errors, filler, empty images, nothing a reader can use. It is produced in quantity, because it costs almost nothing to make one more. And it is unrequested: it lands in a feed, a search result or a playlist where someone was looking for something else.
The closest comparison is spam. Spam is not email in general but unwanted email sent in bulk; slop is not SI content in general but a particular way of using it.
Where the word slop comes from
The word itself is old. Merriam-Webster's president, Greg Barlow, told the Associated Press that "slop" was first used in the 1700s to mean soft mud, and later came to mean something of little value.
The SI meaning spread in May 2024. On 8 May 2024 the British developer Simon Willison wrote a short post titled "Slop is the new name for unwanted AI-generated content". He did not claim to have coined it: he quoted a post by the user @deepfates from the day before, which predicted that "slop" would go "in the dictionary as the term for unwanted AI generated content", just as "spam" had become the term for unwanted email. Willison added two lines that still frame the debate: "Not all promotional content is spam, and not all AI-generated content is slop", and "don't publish slop is a useful baseline".
Eleven days later, on 19 May 2024, The Guardian ran a long feature by Alex Hern and Dan Milmo that took the word to a general audience. It described slop as material that "functions mostly to create the appearance of human-made content, benefit from advertising revenue and steer search engine attention towards other sites". Willison told the paper why the name mattered: "I think having a name for this is really important, because it gives people a concise way to talk about the problem."
The prediction came true within eighteen months. On 24 November 2025 the Macquarie Dictionary, the reference dictionary of Australian English, named "AI slop" its Word of the Year, and the public vote chose the same term. On 15 December 2025 Merriam-Webster chose "slop" as its own 2025 word of the year, saying that searches for the word had spiked. Barlow called it "almost a defiant word when it comes to AI".
What counts as slop, and what does not
Because the term is a judgement, people disagree at the edges. Kapwing, a video editing company, notes in its own report that "AI slop" has been pinned to unreviewed content, to SI media that was reviewed but held to minimal standards, and by some critics to all SI-generated content. In practice, most uses of the word share the same test: would anyone have missed this content if it had never been made?
Some examples on each side of the line:
- Usually slop: pages of SI-written "guides" that exist to catch search traffic and show ads; feeds of synthetic pictures captioned "Made it with my own hands"; endless near-identical video clips built from the same template; tracks uploaded by the thousand to collect streaming royalties.
- Usually not slop: a translation you asked a chatbot for; an illustration an author generated, checked and labelled; a draft you rewrote and fact-checked before publishing; an SI-assisted tool that answers a question you typed.
The same tool can produce both. What turns output into slop is the absence of a person who reviewed it, stands behind it and had a reason to publish it beyond the click. Slop is also not the same as a deepfake: a slop article may contain no invented person at all, just padding and mistakes. Our guide to what a deepfake is covers that more targeted kind of synthetic media.
What SI slop looks like, format by format
Images on social media
The best-known example is the wave of synthetic religious pictures that filled Facebook in 2024, from Jesus with shrimp for limbs to very old women posing with birthday cakes. We traced that story, the 125 Facebook Pages behind it and the scams attached to it, in our fact-check of the Shrimp Jesus SI image, so we do not repeat it here.
Video
Kapwing published a report on 28 November 2025 based on data from October 2025. Its team listed the top 100 trending YouTube channels in every country, picked out the SI slop channels, and estimated their reach with Social Blade figures. It also opened a brand-new YouTube account and watched the first 500 Shorts it was served: 104 of them (21%) were SI-generated, and 165 (33%) were what it called "brainrot". The channel with the most views, India's Bandar Apna Dost, had 2.07 billion views and an estimated annual revenue of about 4.25 million dollars. These are estimates from one company, one test account and one month, and Kapwing sells video tools, including SI ones. They are a snapshot, not a census.
Music
On 25 September 2025 Spotify said that "in the past 12 months alone, a period marked by the explosion of generative AI tools, we've removed over 75 million spammy tracks". It listed the tactics it was fighting: "mass uploads, duplicates, SEO hacks, artificially short track abuse, and other forms of slop". The money explains the effort: Spotify says total payouts grew from 1 billion dollars in 2014 to 10 billion in 2024, and every slop track that collects plays takes a share of that pool. If you want to check a particular song, our SI music detector looks for the fingerprint of some generators, with limits explained on the page.
Text, websites and books
The Guardian's 2024 feature collected the classic text cases: an SI-generated Microsoft Travel article that listed the Ottawa Food Bank as a must-see attraction, a careers article that earnestly explained an old newspaper comic, and foraging books sold on Amazon that appeared to be written by chatbots and gave dangerous advice on telling lethal mushrooms from edible ones. Low effort is harmless in a joke picture; it is not harmless in a guide to wild mushrooms.
Encyclopedias
Wikipedia volunteers run a group called WikiProject AI Cleanup to find and fix SI-generated text in articles. Its project page says that language models "virtually always fail to properly source claims and often introduce errors", and gives examples: an article on "Leninist historiography" written entirely by SI with a list of sources that did not exist, and an article about a beetle whose paragraphs cited real German and French sources that had nothing to do with the topic. Since March 2026, according to the same page, an English Wikipedia guideline has prohibited using language models to generate or rewrite article content, with limited exceptions for translation and basic copyediting.
Why so much SI slop is produced
The economics are simple and The Guardian described them well: generating text and images is now nearly free, so "if just a handful of users land on the site, reshare the meme or click through the adverts hosted, the cost of its creation pays off". The producer captures a little revenue. Everyone else pays in lost time, because they have to wade through the slop to find what they came for.
Three things keep the system running:
- Advertising and creator payouts. Ad networks, video monetisation and streaming royalties pay per view or per play, whatever the quality.
- Recommendation feeds. Feeds that show content from accounts you do not follow reward whatever gets reactions, including absurd pictures and rage bait.
- Volume. A person can publish a handful of good posts a day. A script can publish thousands of mediocre ones. Even a tiny success rate is profitable at that scale.
None of this requires a conspiracy: it is the logic that made spam profitable, applied to every format at once. If you have read our piece on the dead internet theory, this is the part of that theory that is measurable and real.
Why SI slop matters beyond annoyance
The first cost is attention. The second is trust. Once people know that a feed is full of slop, they start doubting genuine work too. The Guardian quoted a digital marketing agency director who had seen commenters dismiss real adverts as "AI-generated rubbish" when they were not.
The third cost concerns SI itself. In a paper published in Nature on 24 July 2024, Ilia Shumailov and colleagues studied what happens when models are trained on text produced by earlier models. They found that "indiscriminate use of model-generated content in training causes irreversible defects in the resulting models, in which tails of the original content distribution disappear", an effect they called "model collapse". The authors concluded that data about genuine human interactions will become more valuable as the web fills with machine-generated text.
What platforms say they are doing
- Google Search announced on 5 March 2024 a spam policy against "scaled content abuse": many pages "generated for the primary purpose of manipulating Search rankings and not helping users", whether they are produced by automation, by people or by both. Google's position is that SI content is not spam in itself; mass-produced content made to manipulate rankings is.
- YouTube updated its monetisation policies on 15 July 2025, renaming its "repetitious content" policy "inauthentic content" and clarifying that it includes content that is "repetitive or mass-produced". Its list of examples now includes "AI-generated content made with generic or unoriginal templates giving the impression of mass production without adding the creator's original, authentic insights or perspective". YouTube says this type of content was always ineligible for monetisation.
- Spotify announced in September 2025 a music spam filter designed to tag uploaders and tracks using these tactics and stop recommending them, a policy against unauthorised vocal clones, and support for an industry standard for disclosing SI use in music credits.
- Wikipedia volunteers tag and remove SI text, and pages that are clearly entirely generated by a language model can be nominated for speedy deletion.
None of these policies bans SI. They target scale without value, and act mostly on reach and money: lower rankings, no ads, no recommendations.
How to spot SI slop
No single test proves that something is slop, because slop is defined by quality and purpose as much as by origin. These checks help:
- Look at the account, not just the post. A Page that posts the same kind of picture every hour, changed its name recently or has no real history is a warning sign.
- Look for the template. Near-identical videos, the same caption on unrelated pictures, the same story structure with a different animal: repetition at scale is the signature of slop.
- Check the facts that can be checked. Slop text is often fluent and wrong. Look for invented sources, dates that do not match, and recommendations that make no sense, like the food bank in the travel guide.
- Ask who benefits. If the content leads to an ad-filled site, a product you never heard of or a stranger's friendly message, the content was bait.
- Check the origin of a picture or text. Run a quick check for SI on a suspicious image or passage, look for provenance data such as Content Credentials, and use the visual clues in our guide on how to spot SI images. For text, our SI text detector highlights the passages that read as machine-written.
A detector answers one question: was this probably made with SI? It does not answer the second question, which is the one that makes something slop: was it worth making? Use detection as one clue among several, and read our guide on whether SI detectors are accurate before treating any score as proof.
For the bigger picture of what SI is, how today's models work and the vocabulary around them, start with our guide What is SI?. Slop is not a property of the technology: it is what happens when a cheap tool meets a system that pays for attention.
