Lately I keep running into the term "AI slop." It describes how AI has made churning out content so easy that the world is now flooded with useless, low-quality stuff. And it doesn't look like a passing buzzword, either. Merriam-Webster picked "slop" as its 2025 Word of the Year, citing how the term spread alongside generative AI.1
Full disclosure first: I'm someone who uses AI to make content. So I know this post could read like self-defense. I still wanted to think it through.
I sympathize with the criticism to a degree. At the same time, I keep wondering, "Is this really a new problem we've never had before?" This post is my attempt to sort out my thoughts between those two feelings.
Most content was always useless
Let's rewind to before AI for a moment. I think most of the content that has ever been produced never left much of a mark to begin with. The sci-fi writer Theodore Sturgeon said back in the 1950s that "ninety percent of everything is crud," and the fact that it's now known as Sturgeon's Law suggests I'm not alone in thinking this.
The worry about a flood of low-quality content is pretty old, too. In the 16th century, as the printing press spread, the scholar Conrad Gessner complained that there were too many books to keep track of. In 19th-century Britain, cheap serialized "penny dreadfuls" were accused of corrupting young people. More recently we had blog saturation, cookie-cutter YouTube channels, and patchwork articles written purely to rank in search.
I'm not trying to blame the people who make things. For content to succeed, a lot of people have to relate to it, want to engage with it, and feel it actually helped them. But creators easily get absorbed in the story they want to tell, so the share of content that meets those conditions has always been low, regardless of the era.
The obvious counterargument
Back then, production cost money, so volume was naturally limited. Now the cost is nearly zero, so it's not the ratio that exploded but the absolute amount.
That's a fair point. But the increase in volume itself has repeated every time the medium changed, and if you look at what happened each time, there's room to see it a little differently. I'll come back to this shortly.
AI opened a door for people who couldn't make things before
So what did AI actually change? As I see it, the biggest shift is that people who used to find it hard to create content, or who needed a long time to do it, can now make it quickly and easily.
Take Instagram comics (what we call "Instatoons" in Korea). There must have been plenty of people with a great eye for funny moments in everyday life who never even got started because they couldn't draw. Same with YouTube. Now those people can put their ideas out into the world and test them.
Something similar happened when smartphone cameras and YouTube lowered the barriers of equipment and broadcast networks. Low-quality video multiplied, but creators who would never have gotten a shot in traditional broadcasting showed up alongside it. I believe that when the number of experiments goes up, the odds of good ideas being discovered go up with them.
The obvious counterargument
More experiments don't help if the good ones get buried. The cost of sifting out what's worthwhile ends up on the viewer.
I agree this is the heart of the AI slop debate. So the next part is about that "cost shouldered by the viewer."
The cost shouldered by viewers is a pattern that repeats every time the content industry grows
The biggest reason AI slop is called a problem is probably that the damage lands on the people consuming content. The more there is to watch, the more the work of finding the good stuff falls on the viewer. I think this phenomenon is real. But rather than something that first appeared because of AI, I see it as part of a macro pattern that repeats every time the content industry expands.
Looking back, every time supply surged, the burden of choosing grew first, and then filtering mechanisms followed.
- When books poured out, book reviews, bestseller lists, and library classification systems took hold.
- When web pages exploded, portal directories came first, then search engines.
- When video overflowed, subscriptions and recommendation algorithms took over that role.
I think the same thing is happening right now. Platform algorithms keep changing so that content people actually watch for a long time, save, and share gets more visibility. Content that gets no reaction rarely reaches most people's feeds, no matter how much of it gets uploaded.
The obvious counterargument
Algorithms amplify what gets the "biggest reaction," not what's "good." Sensational, clickbait content can actually spread better.
True. What the algorithm selects for isn't the same as quality, and that gap was a problem well before AI. Still, I think the overall direction holds: mass-produced content that can't even get a reaction is increasingly filtered out. The filter isn't perfect, but it keeps evolving.
The unfamiliarity of "made with AI" makes the problem look bigger
So why does this problem feel so outsized right now? I think it's because making content with AI isn't yet a familiar concept to us.
Right now, when we see a piece of content, the first thing we notice is "this was made with AI." That judgment slips in as a first layer before we evaluate the content itself. And underneath it there's often an assumption: "if it was made with AI, it must be low quality." There's even research showing that the same artwork gets rated less favorably when people are told it was made by AI rather than by a human.2
I see a parallel with fine dust (air pollution). Plenty of people check the daily fine-dust index and feel the air is getting steadily worse, but Seoul's PM10 levels have, with ups and downs, generally trended downward since measurements began in 1995. In 2010 the annual average for the Seoul metropolitan area was 47㎍/㎥, 32% lower than in 2003 and the lowest since records began.3 The point is that once something gets a name and starts being measured, our perception of it can grow.
Over time, I think making content with AI will feel as natural as using plastic containers. When we're served food today, we look at whether it tastes good, not at whether the dish is plastic. Won't AI become a tool like that, where we look at "what's in it" rather than "what it was made with"?
The obvious counterargument
Fine dust actually decreased, but AI content has actually increased. And plastic didn't just become familiar; it left us with the problem of microplastics.
Both points are valid. So the fine-dust analogy isn't meant as evidence that "things actually got better," only that naming and measuring amplify perception. The plastic analogy likewise doesn't mean every problem disappears once we get used to it. It's closer to saying that only after we get used to something can we see more precisely what the real problem is.
What still deserves our guard: use that betrays expectations
None of this means there's nothing wrong with using AI. There are clearly areas where our society expects AI not to be used.
Resumes are a good example. Using AI to help organize your own career and experience seems like a perfectly fine use. But pasting in a job posting and saying "write me a resume that will get hired here" is different. The reader wants to see this person's actual experience and thinking, and that use deceives that expectation.
Fake reviews, articles fronted by experts who don't exist, and mass posts disguised as human-written fall in the same category. These are problems not because their quality is low, but because they destroy trust. So inside the single term "AI slop," merely clumsy content and content that deceives the viewer are mixed together, and I think the latter is what really deserves our guard. Among the things being called slop, there are surely many cases where the creator was genuinely trying to make something good.
The obvious counterargument
Good intentions don't matter if the result is harmful. There's research showing that repeatedly training AI on AI-generated data makes models lose diversity over time.
Indeed, a 2024 study published in Nature reported this "model collapse" phenomenon.4 Separate from the quality visible to human eyes, the possibility that the entire information ecosystem could get polluted is a clear point of difference from past content floods. I want to leave this as an open question that optimism alone can't settle.
An era that's getting harder for creators, and not all bad for viewers
Once making content with AI becomes the norm, making it will no longer be much of a competitive advantage. Anyone can make it, so from the creator's side, the difficulty of standing out will keep rising. I feel this is already happening, and as someone in the same line of work, it's not easy for me either.
But if you ask whether this change is moving only in a direction that hurts viewers, I lean toward cautious optimism. Attempts are increasing, filters are evolving, and the more familiar it gets, the more we'll judge content by "is this useful to me?" rather than "is this AI or not?"
The obvious counterargument
That optimism is premature. Until the filters catch up, the fatigue and the cost of deception that viewers pay are real.
Agreed. That's why, rather than treating AI slop as a massive new social problem that never existed before, I see it as a phenomenon that shows up in the process of content becoming more diverse, while not brushing off its costs. As a creator, there's one standard I want to hold onto.
Who reads this content, and what does that person actually walk away with?
If I can answer that question, then whether or not I used AI, I don't think it's slop.
각주
-
AP, "Merriam-Webster's 2025 word of the year is 'slop'" — ABC7 ↩
-
Bellaiche et al. (2023), "Humans versus AI: whether and why we prefer human-created compared to AI-created artwork," Cognitive Research: Principles and Implications — PsyPost summary ↩
-
Sanup Ilbo (Korean industry newspaper), "The air in the Seoul metropolitan area has gotten better" (2011) — original article (Korean) ↩
-
Shumailov, I. et al. (2024), "AI models collapse when trained on recursively generated data," Nature ↩
