AI detectors are not just useful, they are actively misleading us into distrusting real writers while missing AI slop completely.
I was sitting on my couch on a fateful Sunday evening, headphones blasting music directly into my ears as I binge-read my saved articles. Not because I didn't have a pile of things to get done, but because I could convince myself I was being productive, educating myself, rather than mindlessly doom-scrolling TikTok or watching Instagram reels.
When I arrived at my ninth article, I found myself zoning out paragraph after paragraph. I didn't want to blame my dissociation on my obliterated attention span—I was stronger than that. But something about that article felt like I'd read it before and was reading it again, but this time styled in a completely different font. Something felt off for sure; I just couldn't place my finger on it. I read that same paragraph seven more times, trying to figure out what I was missing.
Then the puzzle pieces strategically fell into place. I realised that the past five articles I had read all had one thing in common—they sounded like they were written by the same author. Similar word choice, similar tones, perfectly poetic, vague experiences, perfect pacing, similar structures. For good measure, I went back to check if I somehow found myself binge-reading articles from the same author, but no, they were written by different authors and published in different publications, yet they sounded so similar.
Then an idea popped into my head , and I did the unthinkable; I copied the first article and pasted it into an AI detector. The first scan’s results told me that the article was 74% percent human-written, 19.6% AI-assisted , and 6.4% AI-generated. I stared at those results for a while, feeling like I'd just betrayed the trust of a writer, like a pervert who spied on someone while they were naked.
But an intrusive thought told me to run the same article through another AI detector , and the results left me confused, because they told a completely different tale. The scan said the article was mostly AI-generated and partly human-written. The third AI detector required me to pay to get the results, and I was most definitely NOT going to pay a whole dollar simply because of my inability to mind my business.
But now I was left with two results that left me far more confused than understanding Picasso's art. So I tested two more articles. Going by ZeroGPT's results—which people claim to be mostly accurate—they turned out to be mostly AI-written with a touch of human creativity. This all happened before I found out about Substack's partnership with Pangram to detect AI-written articles on Substack to "build trust" and "transparency" between authors and readers.
Apparently, we have been consuming AI slop without even realising it.
I don't personally know the creative process of any writer on Substack, but I know what my writing process looks like. It looks like me sitting at the edge of my bathroom jacuzzi, fully clothed, enjoying the silence, no quiet hum of the AC, just me, my headphones blasting with music and my notes app.
Who am I to point fingers simply because a wonky AI detector says some of my favorite reads were AI-generated? As you already know , curiosity killed the cat (me), so I ran my very own article through ZeroGPT, maybe to prove a point. A part of me already concluded that my results would be different because maybe these writers used AI to write; I could not be too sure. Guess what the results said? Drum roll please...68.3% human-written and 31.7 AI-written. I stared at those results, wondering where I had gone wrong in life.
AI written????
AI written????
Suddenly, I was looking at my very own work from the lens of an imposter. The words were mine, the time was mine, the nights I stayed up at three in the morning because I write better when the world is quiet, those were also mine. But ZeroGPT 's results told a different tale.
I refused to believe it, and I didn't quite understand how these AI detectors worked—until now.
AI detectors only tell you what is most likely written by AI, practically predicting AI writing. After removing some em dashes, ZeroGPT said my article was 98% human-written.
One thing is certain: these AI detectors aren't always accurate, because if you can only detect AI writing by structure, by a certain punctuation, and word choice, then it's bound to fail.
So how can we truly tell the difference between AI slop and authentic human writing?
My hypothesis: we produce what we consume. Read enough AI slop , and you subconsciously learn to write like an LLM—same structures, same safe wording, same predictable rhythm. An article can be 100% human and still read like a magicbot wrote it, because that's what the algorithm rewards. And yet we're supposed to trust detectors that flag writing style alone? Style that any human can mimic?
I have seen so many takes on Substack's partnership with Pangram. Some people believe it does more harm than good to the trust between readers and writers. Others believe Pangram is stealing our writing and using it to train AI models, while many believe it's a good move Substack has made.
These could all be true.
Substack's partnership with Pangram leaves me conflicted. Yes, Substack—just like every other social media platform—is slowly becoming saturated with AI content. Do I want to read AI slop? No.
But what about the writer whose career collapses because an AI detector inaccurately flagged their work as AI-generated? It's not just about trust; it's about fear.
I read in an article where a writer said the real danger writers face is not having our work stolen by Pangram and used to train AI, but having people copy and paste articles on Substack into ChatGPT to generate well-articulated comments to gain favours from the algorithm.
I have thought about this for a while now. The real threat we writers face isn't Pangram stealing our work and using it to train LLMs, but the writers who plagiarize other writers' work by asking Claude to rewrite it.
Telling the difference between AI writing and human writing can be tricky at times. One thing I have noticed is the vagueness in details, especially in personal stories or reflective pieces. LLMs tend to avoid sensory details because they can't write about something they've never lived. They just compound experiences they are fed from the Internet,and that's how we end up with the cliché "lived experiences" and the predictable metaphors recycled across hundreds of Substack articles.
Take this passage from one of my own articles:
I still remember my fifteen-year-old self standing in the hallway, peeking from the corner of my eye when a scene from my favorite telenovela took a dangerously wild turn. Trying and failing to look anywhere except the TV. Suddenly the walls looked stunning. The chandelier caught the light with a glint I'd never noticed before. I sank deeper into the couch cushions, hoping they'd conceal me from my mother's eyes.
But I watched. The walls listened—to the moans, the merging of bodies. I committed every detail to memory, not because I wanted a how-to-kiss-like-you-mean-it degree from the Institute Telemundo, but because I thought that was the highest form of intimacy. The purest form of love in all its animalistic, primal glory.
This shows a real lived experience no LLM can replicate because it's yours.
In a world saturated with sexual imagery and hyper-romanticized connection, we have accidentally starved ourselves of something far more essential: the quiet, unassuming power of non-sexual intimacy. We have conflated closeness with carnality, forgetting that the human body speaks in many dialects. The soft press of a hand, the absent-minded stroke of a back, the shared silence of a morning coffee—these are not foreplay; they are the very fabric of human connection.
This paragraph is beautifully written, but it's missing something—the human touch. A lived experience. One makes you feel something. The other makes you admire how well it's written. One has a body. The other has a polish.
Authentic human writing feels like a writer baring their thoughts to the reader. I call it thinking (bleeding) on paper. We constantly produce thoughts and as a writer, sometimes you struggle to arrange those jumbled thoughts in a way that doesn't sound incoherent, but instead like an intimate ritual between the writer and the reader.
AI slop, on the other hand, sounds too composed. No straying thoughts begging to be liberated before other thoughts. Every sentence ends with a resolution, poetic one-liners that leave no question for the reader to actually sit with their thoughts.
Sometimes I can't even explain it. Sometimes I may be wrong.
I find myself sitting on my couch for the third day, writing this article , and I have come to a conclusion. The real threat we face is not just AI writing and paying big companies monthly subscriptions to tell us the difference between AI writing and human writing, but the fact that we are losing the ability to recognize each other's words without the help of a magic chatbot telling us what's real and what isn't.
No posts
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.