AI Slop Isn’t
AI slop is a silly term, and we should abandon it
TL;DR: Two recent papers discuss the utility of AI slop, offering alternate viewpoints. I argue that the conceptual disagreement between the two is a consequence of the term “AI Slop” being, for lack of a better word, “sloppy.” Let me contextualize first.
Kommers et al. (2026) argue that what is called “AI slop” is a legitimate form of expression, and that the bashing is just the latest highbrow attempt to demean “low” art. They correctly note that AI-generated videos serve a social function and have aesthetic value. AI slop pushes the barrier of producible content (at least on a budget), and people can make creative, entertaining content with AI. They also emphasize that AI slop circulates within a “survival-of-the-fittest” algorithmic environment: if some of this content flourishes, they suggest, that fact reveals something about human preferences and collective demand. Yet, when justifying the value of AI slop, their example and arguments heavily prioritize personalization; they write that:
For example, many children would find it entertaining to watch a video of their father dancing wildly in his underwear.
Nichal et al. (2026) take issue with this example, arguing that they represent a charitable and somewhat atypical version of what people usually mean by “AI slop.” For them, the relevant object is not the lovingly crafted, personalized video but mass-produced, AI-enabled, engagement-optimized content. They note that, even if people individually may like AI-generated content, its rise may move us towards an objectively worse information environment, and do so while validating the very controversial “project of AI” and its environmental and political consequences.
My concern is that the discussion, as currently framed, may not be very productive because it rests on an unstable conceptual foundation for what counts as “slop.” Kommers et al. define it as AI content that demonstrates superficial competence (it looks fancy), asymmetric effort (it is easy to make), and mass producibility (it is easy to automate). Each and every one of these traits is highly questionable as a way to describe the generative AI content that most people encounter on social media these days. (And dare I say, in general).
These definitions rely on fast-moving targets. Technology has historically made it easier to produce “content” that looks “better” with less effort (everyone has an amazing camera in their pockets these days). Filters, editing apps, stock music, and auto-captioning have all lowered the effort required to produce content that looks competent. Yet we do not usually treat CapCut edits as “slop.” Just think about how we changed how we evaluate written text. If a time traveler came back to 2019 with GPT 5.4, they could probably become an award-winning writer (yes, it would!). But we devalued AI-generated prose once it became widely available — even I, who am quite happy to use ChatGPT as a writing assistant, despise the “it’s not about X, but about Y” constructions that pop up all the time
Thus, “superficial competence” and “asymmetric effort” are only superficial and asymmetric given a specific moment in time. Given another baseline, all of TikTok since its inception would be described as such, regardless of AI. But mass-producibility is not a great descriptor of AI slop either. The pre-slop era is filled with channels with some mechanized voice reading a Reddit story while someone very skilled plays Subway Surfers. Or a quick narration of an older B-roll movie in 12 parts. Or very low-effort reaction to other content.
Tiktok failed to load.Enable 3rd party cookies or use another browser
AI-generated content captures people's imagination in a few ways: first, by showcasing the capabilities of new models. For example, when Veo3 was unveiled, we saw an influx of natural disasters and Cat Olympics. Second, AI-generated content has become a medium for memetic play. Italian brainrot, for example, was a collective world-building exercise where people invented ridiculous characters, each with funny catchphrases and obscure backstories (see the Wiki). Fruit soap operas work in a similar way: the humor comes not only from the fact that fruit has been anthropomorphized, but from the melodramatic excess, the repetition, and the absurd tropes. Third, AI-generated content often has its values in the context in which it is shared! Think White House ICE memes, or the recent Christ the Redeemer vs. Statue of Liberty brawl posted by Iranian official accounts (the latter stood no chance, as Christ the Redeemer is clearly into BJJ).
Tiktok failed to load.Enable 3rd party cookies or use another browser
Last, there is content that is simply engagement farming. Like this account that only posts a baby dancing (to be honest, it is kind of impressively well made). But here is where Kommers et al. are right: there is a visceral distaste to AI-generated content! The more brutal reality is that the vast majority of short-form video content adds very little to people’s lives. There are literally dozens of accounts of people smashing things on a hydraulic press, with millions of views. The AI Slop discourse creates this line between high-quality, thoughtful content and “slop” that makes little sense on the short-form video platforms where this content flourishes. Still, creating successful content online is freaking hard --- I dare academics to try (I did), just to be humbled! Even though these videos seem silly, their success depends on deep understanding of the zeitgeist of short-form content.
Tiktok failed to load.Enable 3rd party cookies or use another browser
Thus, let’s be done with the term “AI slop.” It is another case where we ascribe too much independent power to technology itself, much like earlier debates that treated “the algorithm” as a mysterious force rather than a name for specific incentives, designs, and human behaviors. Generative AI did not invent content farming nor the strange pleasures of short-form video. The way in which it changed the incentives to producing such content matters, but “AI slop” is the wrong name for the problem. It collapses too many different phenomena into a single sneer. It is, dare I say, sloppy.
I liked that you articulate what goes into calling something slop. I think there’s a lot of social dynamics behind it too, it’s not only about quality, it’s also about social status, and wanting to differentiate yourself from what becomes commodity
Agreed. We’d all be doing ourselves a favor by doing so.