At the risk of sounding like a teenager kicking off a torturous essay, Merriam-Webster defines slop as:
digital content of low quality that is produced usually in quantity by means of artificial intelligence
So far, so concise. But I don’t think this definition really does the term justice. I’ve been writing about slop for a couple of years without really defining it myself in articles like Digital plastic in 2023, The Effort Economy of Slop in March, and AI;DR in July. The closest I came to a definition in any of those posts was lifted from a Bluesky user, and shared by software engineer and writer Simon Willison:
Slop is something that takes more human effort to consume than it took to produce.
I prefer this “effort economy” definition, because it speaks to the imbalance between author/reader. From people’s reactions to AI slop online, there’s definitely a sense of frustration, and of broken trust. But what else is there?
Many of the readers of this blog will know that I spend far too much time on LinkedIn. That platform has recently added a whole button for reporting AI slop – though it’s clearly a fairly cynical attempt to gather training data for their own AI detection models.
Anyway, I turned to the audience on LinkedIn to crowd source a definition. I ran three surveys over consecutive weeks, each built from the results, the comments, and the ideas of the one before. Around 450 votes and more than fifty comments later, I think there’s a better definition in there.
Question one: AI slop is…
The first poll tested a few ideas about where AI slop is located, and the opinions towards it. Is slop a property of the writer or the reader? Is it a price we have to pay because of the technology? Something “in the eye of the beholder” and based on reader interpretation?
Just over half of the 188 respondents put the fault in the author. Only 8% were willing to label slop as an unavoidable by-product of the technology, which means the technology itself got almost no blame at all. Slop, on this reading, isn’t something AI produces; it’s a human byproduct of using the technology without due dilligence.
That stands a little at odds with the Merrian-Webster definition, which centres the technology.
17% also went straight to unforgivable, suggesting a moral category rather than an aesthetic one. In the comments, Sarah-Jane B argued that how offensive you find slop follows how sensitive you are to quality in that domain, writing that “slop is the kindy recorder class of sound.” My suspicion is that the people with the sharpest slop radar (slopdar?) are the ones with the most exposure to the outputs rather than the best aesthetic ear, though Sarah-Jane’s follow-up split it neatly into a fast, intuitive read and a slow, deliberate one, with a third group not thinking about it at all.
Johan Cedmar-Brandstedt, who has been reporting AI slop since before it was cool, shared the Merriam-Webster definition, and disagreed with my argument that slop isn’t necessarily “produced in quantity”. To me, the dictionary definition suggests the kind of article mills and clickbait farms we’ve seen online for decades, whereas I see slop at a much more personal level, with people generating their own “unique” slop on demand in LinkedIn comments and Substacks. Johan disagreed on the premise that every use of an LLM is basically a transaction between a user and a platform.
Two comments pointed towards the second poll. Victoria Spicer-Stuart wrote that “AI makes it very simple to create very lengthy texts very quickly, with little input from the ‘author.’ Show some discernment and think about your audience.” And Michelle P., on why she reacts badly to obvious AI in an email, wrote, “I find the slack or lazy use of AI particularly annoying as I can no longer see a message through the specific person’s lens. It ends up becoming generic messaging and everyone sounds the same.”
Question two: AI slop is bad because…
Slop is certainly divisive. Some people claim they don’t care, as long as the message gets through. Others recoil in horror, disengaging with texts at the merest hint of AI use. Both public and professional attitudes towards AI slop have shifted in recent months as the technology continues to grow in ubiquity. Authors are losing book deals, scientists are embarrassing themselves with vibe coded errors, and, perhaps most alarmingly, AI image slop is threatening to ruin the internet’s primary source of enjoyment: cute animals.
So, why is it so bad?
Despite a focus in education policy on inaccuracy and deception, these two items polled only 22% between them. My LinkedIn audience is mostly educators, but even so the problem with AI slop appears to be less about hallucinations and “truth”, and more about care, effort, and attention.
In the comments, Barend Last wrote:
When it’s unedited and low-quality, it’s a waste of my time and other resources. I don’t want to read it, and I don’t think it should exist. However, 99.9% of that also applies to a regular library. So my main problem is that it’s algorithmically pushed into my timeline, even though I didn’t ask for it.
LeeAnn Lindsey articulated a similar point: “It’s noise. That distracts from more valuable content”, as did James Tsatsaronis, who noted that slop “takes valuable space and attention away from other things that actually matter.” So slop is a bandwidth problem as well: readers have a limited amount of attention to spare on the content they read online, and when that bandwidth is taken up by slop, it makes us feel resentful. A single slop-paragraph in a news article is a minor irritation, and maybe a sign of carelessness. The same paragraph, multiplied a thousandfold by an obscure algorithm that decides what you see on your feed, is a different scale of problem.
Several people asked where the “all of the above” option was. Leaving it off was deliberate, because I suspect a lot of people would have clicked it. As it stands, perceived effort and ubiquity are what carried the vote.
The trouble with effort
The “effort” aspect is more complex than it first appears. Libraries are full of bad books written entirely by humans, and nobody calls them slop, because nobody is standing over you forcing you to read them. In the survey 54% chose low effort, and the comments section spent the rest of the week picking that answer apart.
I’ll declare a conflict of interest here, because effort is part of my own answer too. The effort economy of slop article mentioned above was built on that heuristic borrowed from a Bluesky user via Simon Willison’s blog: if the effort to consume what you’ve made exceeds the effort you put into making it, you’re producing slop. I argued that GenAI inverts the usual asymmetry of communication, dropping the producer’s effort to near zero while the consumer’s stays the same or climbs, and that the cost doesn’t vanish so much as transfer; to the reader, the teacher, and the person downstream who now has to sift through the AI bullshittery to find the meaning behind the words.
Kaywin Cottle, in the comments on the second poll, threw a spanner into the effort-works:
Just because something takes effort doesn’t make it good, valuable, or worthy (not slop). I can write for days and it take lots of effort, but still the final product ends up in the trash. Sometimes, I can jot a quick, low-effort note that I find very valuable in the end.
Michelle P. put the same objection on the third poll and named the missing piece:
…Is there something missing about intellectual responsibility and judgement? That feels bigger than ‘effort’. You can still put a lot of effort into producing slop 😬
Brett Graham, explaining why he didn’t pick “carefully edited”, wrote that “you can carefully edit something that in the end is of no value at all.”
So, “effort” is problematic for a couple of reasons. You can pour hours of AI-free time into something worthless, and you can dash off something worth reading in seconds. You can carefully labour over a piece of AI-assisted prose, and still end up with something that a reader might identify as slop. Or you could, theoretically, generate a thousand off the cuff zingers with an LLM model in minutes and, provided the reader didn’t notice, “get away with it”.
Is it even an AI problem, then? Tim Evans didn’t think so:
I also wonder why this becomes fundamentally different because AI is involved? We’ve never really known the effort or process behind what we read. Maybe AI just makes that uncertainty more visible.
I think he’s right, and it’s the same problem I described in AI;DR from the reader’s perspective. The careful hybrid writer and the lazy prompt-paster might get sentenced with the same verdict of AI-slop, because the sophisticated use of these tools happens entirely in the head and browser of the author and doesn’t necessarily reach the finished product. Readers have never had a true insight into the mind of an author: there’s decades of literary theory built around that exact idea.
“Effort”, then, is a proxy: how much labour does the reader think the author put into their work, into sharing their ideas? And since we can’t ever really know the answer to that, and there’s nothing an author can do to truly convince a reader of their process (after all, the author could just be lying…) it will always be biased towards the reader’s opinion.
Question three: AI-assisted writing is not slop when…
Finally, the third poll looked for a negative definition. If AI slop is ubiquitous, low (perceived) effort, and careless, then what does AI-assisted writing look like when it isn’t slop?
The road to hell is paved with good intentions, and it seems that even the best of intentions won’t save you from being accused of AI slop. Back in March I argued that AI slop videos are worse than clickbait, because clickbait at least had intent. I said that slop is production without intent, or communication without a communicator. The poll suggests the reverse is also true: good intent doesn’t redeem the output any more than bad intent condemns it. You can mean well, work hard, and still waste my time, basically. Once again, the reader can’t see motive, effort, or intent: only what they perceive of it.
James Dillon chose “careful editing” and explained in the comments that “The human gaze is still required. When pure agentic content no longer requires a human editor… that will be a fascinating next phase.” Brett Graham chose the runner-up – “it says something interesting” – for the opposite reason, that careful editing can’t rescue an empty argument.
Dave Cormier wanted a fourth option, labelling, and argued the case for it in terms of identity rather than compliance:
I value identity. Who thought the thing. What is the context in which the thing was thought/said/written. If I got to choose, all ‘generated’ text would be marked as such, and any human contribution would as well. Then I know more about what context to understand it’s meaning.
David Brown asked a question about the poll itself: “What does ‘AI-assisted’ mean? Is Grammarly ok? What if you genuinely use ChatGPT to check your grammar instead? Is that OK? What if you use AI to generate a transcript?” I personally use LLMs to turn rough voice transcripts into HTML, and I have no idea where on his scale that sits, so there’s definitely some contested territory over what “AI-assisted” actually means.
Danielle M. arrived at a three-part test close to where the polls collectively land, identifying slop at the intersection of “intent, quality and volume”, with a distinction between AI content that’s poor because someone was rushed, and slop proper, which is bulk and often bad-faith. Dwayne Ripley pointed at one of my favourite 2026 terms, meat proxy, for the person who passes along machine output without reading it.
And finally, a couple of people declined the exercise entirely. Outi Leppänen wrote that “None of those [options] wash it clean,” and Anke al-Bataineh stated that “There is no ethical usage of LLMs.”
The definition
Putting the three polls together, the fault is largely human, not technical. The damage – reputational, attentional, interpersonal – is a transferred cost that’s multiplied by volume. The only apparent remedy is a transparent, effortful use of the technology, or no use at all. And the ultimate verdict belongs almost entirely to the reader, who can’t see the process or the inside of the author’s mind, but nonetheless casts the final judgement of whether the media is AI slop.
slop /slɒp/ noun
1. Digital content that takes more effort to consume than the reader believes it took to produce.
2. Obviously AI-generated content, especially as distributed algorithmically to readers who did not ask for it.
3. derogatory. Content produced by an author perceived to have outsourced judgement to an AI model and published the output unedited or unread.
Usage: Slop is a judgement, not automatically a property of the text. Because a reader cannot observe an author’s process, effort, or intent, the label attaches to what the work appears to be rather than to how it was made. Careful hybrid writing and lazy prompt-pasting can both be perceived as slop.
I wish people would stop leaving AI slop in my LinkedIn comments!
I was reading the article, and about half way through I realised it was mostly slop.
Broken contracts
Like I wrote in the earlier AI;DR article, there’s a sense of broken, invisible contracts between the reader and the writer. The old author/reader contract, the assumption that a real person was here, doesn’t fly any more. No amount of detection tools, watermarking, or AI declarations is going to repair that contract. All the author has left to stand by is their word.
Perhaps AI slop is a temporary word; part of the zeitgeist and AI bubble we find ourselves living through. We need a word for what comes out the other end when the cost of producing text collapses, and when the reader spots something that, to them at least, doesn’t uphold the contract.
Thanks to everyone who voted and argued in the comments across the three weeks. If you’d have asked me four years ago whether I’d be attempting definitions of new words based on LinkedIn polls, I probably would have laughed myself off my chair.
Want to learn more about GenAI professional development and advisory services, or just have questions or comments? Get in touch:
Leave a Reply