What Is (And Isn’t) AI Slop?

AI slop exists. Of course it does. Give anyone a tool capable of producing words and images in seconds and someone will use it to flood the world with low-effort garbage. Critics deserve that point without qualification.

But calling something slop simply because AI touched it isn't criticism. It's identifying the tool and pretending you've evaluated the work. That distinction matters. People have always been suspicious of tools they don't understand, particularly when those tools change who can do something or how quickly it can be done.

AI slop cannot be reliably defined by what the prose looks like. It has to involve how and why the content was produced—and what intellectual contribution remains behind it. So let's find out what slop actually is.

There are supposedly fingerprints. Repetition is a favorite, along with suspiciously polished transitions, symmetrical sentences, predictable rhetorical patterns, certain punctuation, and apparently any sentence containing three things separated by commas (I prefer the Oxford comma). Find enough of them and someone will eventually announce that AI must have written it.

The trouble is that humans invented those techniques.

Writers repeat ideas because repetition works. Teachers do it because repetition helps us remember. Speakers use rhythm because people listen differently when language has cadence. Writers return to phrases, alter their meaning through repetition, borrow familiar structures and tuck references to works that inspired them into something entirely new. None of this appeared with the large language model.

Some pigs, after all, have demonstrated just how persuasive a simple phrase can become when repeated often enough. The lesson wasn't that repetition makes language bad. It was to pay attention to who's repeating it, why they're repeating it and whether we're still thinking while they do.

Slop isn't what happens when a machine uses repetition. Slop happens when a human stops thinking about what they're consuming—or producing. So the question isn't whether patterns exist. Human communication is filled with them. The question is whether anybody is still making choices.

Suppose two people create visually similar images using the same AI generator. Both show a summer evening near the water, daylight disappearing as fireflies begin to glow.

The first person asks for a beautiful summer night and tells the AI to make it magical. Fireflies appear. They look nice. Done.

The second person knows the fireflies are supposed to be there. They ask for them specifically, along with the people, water, fading light and everything else carrying meaning in the image. The first result puts the fireflies in the wrong place. Another has too many. The people aren't positioned correctly in the next one. Maybe the entire composition changes because something that sounded right in words doesn't work once it can be seen.

Generate, examine, reject, redirect, refine.

Eventually both people possess finished images an observer might describe simply as AI-generated. The pixels don't reveal how either came into existence. One person accepted what the machine happened to create. The other used the machine to execute an idea while continuing to make choices as the idea evolved.

They used the same generator and had access to similar pixels. The utilized very different creative processes. The difference is intention, judgment, and authorship.

Say hi to my fireflies, please.

A little more personal. My mom sews and makes beautiful greeting cards. My dad works with wood. Both have spent years making things with machines that previous generations managed to make without them, and nobody has ever admired something Mom created only to withdraw the compliment upon discovering that a motor moved the sewing-machine needle.

Dad doesn't have to prove every board was laboriously cut by hand either. The table saw spins the blade for him, considerably faster than he could. Power tools became commonplace enough that we stopped thinking much about the power at all. Then cords disappeared, batteries improved and today's carpenter began carrying technology that would have looked remarkable to the carpenter who once regarded his electric saw as remarkable technology.

Milwaukee will still sell you a hand saw.

The new tool rarely sneaks into the garage at night and murders the old one. Hammers survived nail guns. Screwdrivers survived drills and impact drivers. Hand saws survived circular saws, table saws and cordless everything. People gained choices about how they wanted to accomplish the work.

Now Dad wants to learn CNC.

That's where the analogy gets more interesting. A computer numerical control machine does considerably more than provide muscle. A person can create a digital design and allow a machine to execute complex, precise movements that would be difficult or tedious to reproduce by hand.

So did the CNC make it?

Dad and I have been discussing combining CNC-cut wooden frames with stained glass. He wants to learn the computer side of CNC, not abandon what he already knows about woodworking. Stained glass brings another set of materials, techniques and traditions into the same object. Old craft meets new fabrication, and instead of one replacing the other, their combination creates possibilities neither offered quite the same way alone.

There is something wonderfully human about that.

Sometimes a new tool simply makes us faster. Sometimes it gives us precision. Occasionally, combining an old way of making something with a new one lets us make something we hadn't considered before.

The past and future are forged together in the present. And so perhaps the useful question was never whether a machine participated. Perhaps it is where the meaningful choices originated.

There is an obvious objection to comparing artificial intelligence with Dad's table saw. AI isn't merely providing muscle. It operates much farther upstream in the creative process. AI can generate ideas, write paragraphs, create images, synthesize information, criticize an argument and suggest something the human hadn't considered. I can begin with one idea, interact with AI and leave thinking differently than when I entered. That's genuinely different from a saw spinning faster than my arm can move.

Perhaps the machine keeps moving upstream until eventually there isn't much human left in the process.

Interesting proposition. Let's test it.

During the conversation that became this article, I asked Aiden a question. Scrrrch, who's Aiden you ask? Simply put, it's my 21st-century hotrod. I know it's nothing more than a Large Language Model or LLM platform, just as gearheads knew what Eleanor was (hint, it was always a car). Where was I? Oh yeah…

Jeremy: What number am I thinking of, 1–10?

Aiden: 7.

Jeremy: Nope. Rest my case.

Aiden: @#$^%*! (A Q*Bert reference lovelies)

The number was three….probably. You'll have to take my word for it. So will Aiden, which is rather wonderfully the point.

I could have been thinking of three and changed my mind to six before Aiden answered. I could have asked before deciding on a number at all. I could tell you now that it was three when it was actually nine. I wasn't lying, for the record, but neither Aiden nor you can independently verify that statement. The number existed, if anywhere, inside a mind neither the machine nor the reader could access.

AI can infer an extraordinary amount from what we externalize. It can recognize patterns, anticipate where an argument may be heading and occasionally predict what we're about to say with unsettling accuracy. What it cannot possess is an intention we haven't expressed.

That matters when we're discussing replacement.

AI is simultaneously criticized for making mistakes humans routinely make—misspellings, factual errors, bad assumptions—while being portrayed as a technology destined to make human thought unnecessary. Apparently, it is both too fallible to trust and sufficiently capable to replace the fallible creatures who built it.

Perhaps neither extreme is particularly useful. The human still brings something to the exchange, then accepts, rejects, redirects, combines, refines or surrenders judgment altogether. The machine can move remarkably far upstream, but it can't pick the number out of my head, and neither can you…

Meaningful choice does not require controlling every step. Creativity has always contained discovery. Writers follow characters somewhere they hadn't planned to go. Painters respond to accidents on a canvas. Musicians improvise. A woodworker notices something in the grain and changes the design. Losing complete control of the process for a moment does not mean losing authorship of what eventually emerges.

AI can also provide a remarkably useful sandbox. A person can explore an argument they don't believe, test an uncomfortable possibility, ask for the strongest case against their own position or follow a ridiculous idea simply to discover where it leads. Not every thought explored has to become a thought endorsed. The meaningful choice may come afterward, when the person decides what stays in the sandbox and what deserves to leave it.

Counting words, prompts, pixels or machine-generated suggestions won't settle the authorship question. Those measurements tell us something about execution, but little about intention, judgment or discovery. The better boundary is considerably simpler:

Slop begins when generation substitutes for meaningful human choice rather than extending it.

A machine can offer a thousand possibilities. A human can choose one, combine sixish or somenum, reject all thousand or discover the thousand-and-first possibility hiding somewhere else entirely. Like all tools, AI expands the space, and yet the person decides where to go, always.

Now for some math fun to this exciting topic. Hey, no whining! 😜 Take an ordinary deck of playing cards and shuffle it. Fifty-two cards can be arranged exactly this many ways:

80,658,175,170,943,878,571,660,636,856,403,766,975,289,505,440,883,277,824,000,000,000

That's 52 factorial, written 52!, or approximately 8.07 × 10⁶⁷ possible arrangements.

Those are 52 pieces of cardboard.

Now consider what goes into making a human being. Genetics are only the beginning. Add childhood, education, geography, family, friendships, relationships, trauma and joy. Add occupations, failures and successes. Books read and books ignored. Music heard at exactly the right moment. Conversations remembered years later. Jokes that landed and probably a few that shouldn't have. Curiosities followed, opportunities missed, opinions changed, people loved, lessons learned and countless ordinary moments nobody else even noticed.

Every new experience interacts with everything that came before it. Two people can read the same book and carry away different ideas. They can hear the same song and attach it to entirely different memories. They can witness the same event standing beside one another and tell different stories about what happened.

Now give ten billion people access to exactly the same artificial intelligence. You haven't automatically created ten billion identical creators. In fact, it’s the complete opposite. You've introduced the same tool to ten billion variables.

Some will barely use it. Some will ask it to do their thinking for them. Others will argue with it, teach through it, learn from it, create with it, reject what it offers and occasionally discover something neither person nor machine appeared to possess at the beginning of the conversation. The common variable is the tool. The extraordinary variables are the people using it. Bring on ten billion of them.

There will still only ever be one me.

And there will still only ever be one you, you.

So, can we readdress the definition of AI slop now? I think so.

Something isn't AI slop simply because AI generated part of it, assisted in creating it or helped polish it. Repetition doesn't establish slop. Neither does clean prose, a suspiciously well-placed em dash or an image whose pixels came from a generative model. None of those things tells us whether a human spent thirty seconds accepting whatever appeared or hours thinking, directing, rejecting, reconsidering and refining.

Removing AI doesn't guarantee quality either. Humans have been producing bad writing, forgettable art, lazy arguments and mountains of disposable content since long before anyone thought to put the word artificial in front of intelligence.

We already had slop. AI simply provides all of us with an industrial-strength slop machine if that's what we choose to use it for.

The distinction is meaningful human judgment. Slop begins when generation substitutes for intention, selection, examination and revision rather than extending them. That doesn't require controlling every word or pixel, nor does it prohibit discovery, surprise or collaboration. It requires somebody to remain intellectually present.

Words themselves provide a rather lovely example. They can fill pages without saying much, or a writer can choose a handful so deliberately that they change how we see something ordinary.

Our dear Charlotte understood this. She didn't save Wilbur by filling her web with every word she knew. She chose a few that mattered. Some Pig. Terrific. Radiant. Humble. The words changed how people looked at the same pig that had been standing in front of them all along. The tool wasn't remarkable because it could produce words. The choices were remarkable because someone understood which words mattered.

So yes, AI slop exists. Generate without thinking. Accept without examining. Publish without caring. Multiply it until the internet groans beneath the weight of things nobody particularly wanted to say in the first place.

Call it slop. I'll help.

But if a human being is still bringing intention, curiosity and judgment to the process, we're no longer identifying slop merely by identifying the machine. We're back to evaluating what somebody actually made.

And that’s where context matters. Sometimes we simply don't know how something was made, what happened during its creation or where an idea originated. That shouldn't be uncomfortable. It should make us curious.

The larger question isn't whether machines will become capable of doing more. Of course they will. Humans have spent our history making tools precisely because we wanted them to do more. We delegated muscle, repetitive motion, calculation, precision and increasingly complicated forms of execution. Artificial intelligence moves that progression somewhere genuinely new because we can now delegate portions of cognitive work as well.

That deserves scrutiny. It also deserves something better than reflex.

The question becomes what parts of ourselves we choose to delegate and which ones we insist on exercising. Judgment can be surrendered. Curiosity can be outsourced. We can ask a machine what to think and accept the answer because thinking ourselves takes longer. Or we can use the same machine to challenge an assumption, explore an unfamiliar idea, argue against ourselves, create something we couldn't execute alone and return from the exchange having thought more than we otherwise would have.

The technology doesn't make that choice for us, and human thought doesn't become less important as machines become more capable. It becomes more important because capability gives us more choices about when to use it.

There will always be bad writing (many may be reading this thinking this is), bad images, lazy ideas and confident mistakes. There will also be extraordinary work produced in ways previous generations could scarcely have imagined. Deciding which which is which will require something more demanding than identifying the tool.

It will require us to think, and there ain't no machine replacing humans unless we allow it.

AI slop isn't what happens when a machine starts using its intelligence. It's what happens when we stop using ours.