Substack is Drowning in AI Slop
I just want to have a conversation
When my grandfather passed away, the only request I had regarding his belongings was a carved bird he made in his woodshop. One of my happiest memories as a youth was exploring that dusty shed and feeling the thick steel of those machines from a bygone era when they were built to last generations. He was not an expert craftsman, as he took up woodcarving in his 60’s as a retirement hobby, but every one of his artifacts told a story of his character. He loved wildlife, an adoration second only to his beloved wife. Both were the main themes of his work.
I could easily go to a store or search on Amazon and find mass produced wooden crafts in my office. Of course this wouldn’t be the same, as the entire joy of having them is that a loved one bared his soul in making it, that human hands crafted it, that a human mind smiled in satisfaction when it was done. He’s been gone for over a decade, but that craft in the office helps me remember him in a way nothing else can. The sculpture became more than a block of wood, but an extension of my grandfather.
Similarly, when one enters a social space, whether online or in person, you want to have an encounter with someone else, an other. By its very nature there are going to be rough edges, little tics that tell more than the words being said, means of communicating that are less than optimal. Every person has a cadence, a stream-of-consciousness that shines through, giving a glimpse into their interior life. Even if he’s role-playing or wearing a mask, some of that humanity seeps through.
When one encounters writing, it’s the same phenomenon. While the written word lacks the physical context and is more refined, it maintains that connection, an interiority shown to the world. Especially in cases of commentary, how one expresses such thoughts is often more important than the information given. In short, the person’s writing becomes an extension of himself, just like any artist who paints or knits or builds. The labor of love makes the text more than just words on the screen. It’s an expression of one’s core being. The Logos, Pathos, and Ethos critical to any writing has to shine, forged from an actual mind who feels these sensations.
While there is a need for dry, impersonal texts that remove this inner realm in law and other formalized realms, those who wish to truly learn and understand a viewpoint have to feel that connection with another human mind. This need for bonding is why AI will never replace human teachers, and why it’s so dangerous for people to think of AI, something that doesn't wonder, dream, touch, or feel pain as a human-like entity.
When you find out someone plagiarized writings or had their work ghostwritten, there’s a sense of betrayal to those who want that connection. Even in a parasocial environment like the Internet, there’s still the assumption of a human being on the other side with the understanding that even if they are hiding as an anon, they are a real person who is making a real effort at connection and understanding. When that work is offloaded, the connection is broken. The inner nuance of their thought is not displayed on the screen, even if their opinions are. With the acceleration of AI writing, now anyone can offload writing to exterior staff, and with it eliminate the inefficiencies of putting thoughts on paper, warts and all.
For a long time, you had to rely on a “sixth sense” to sense AI writing. The tools for detection were easily mitigated and were notorious for lack of consistency. The accusation of AI slop is a serious accusation, and people were loath to toss it out without good evidence. Luckily detectors are getting better, are harder to dodge, and give fewer false positives. While there is still error, it can reinforce the “AI smell” sixth sense. Look at a few articles with the same 100% AI Conclusion, and you are almost certainly dealing with a slop merchant.
To test Pangram, I ran the experiment on my own articles and posts I knew were before A.I, and therefore safe. It ran flawlessly. In my entire SS career, I used AI input in precisely one. In that case, I wrote the entire article and then used AI to go through the draft and change the narrative voice. I’ll leave it as an exercise for the reader to guess which article I’m referring to. Even after significant modifications because the A.I.s output modifications were now that great, Pangram picked it up. Mixed at 50% A.I… well played. Even though it was largely a jokey article, it gave me a bad taste in my mouth and I vowed to never to do it again.
Picking up slop has become easier too, as they now have a plugin feature to quickly ascertain the level of AI., meaning you don’t have to manually input a snippet of text. When I installed it, I expected a few here or there with most in small accounts with no paid subscribers. Needless to say that hopelessly naive thought was brutally dashed, and my frustration and curiosity led to a little experiment.
I took the 20 Top Rising substances on the site as a whole and found a random article to test with Pangram. Note these are Substacks where people are paying to read, and many have thousands of subscribers. The results are below:
Note: There are some obvious slop ones here, but there is the possibility of false positives, especially mixed. YMMV.
Top 20 Rising:
Gad World - Not English, did not check
Dan Djik - Not English, did not check
Pod Street Walk - Podcasts
Kyle | OtterList -100% human (this seems to be a business portal)
Maura Brannigan - Mixed (Maura said it is 100% human in the comments, and I believe her, though looking at other articles, it’s all over the map where most posters are very consistent)
Ezéchiel Zérah - French. Did not check
Iraq Dinar Investment News - All Paid Articles
Rudy & Rooster - Not even text, just stock charts
A lot of these are investment and get-rich-quick articles that feed on gullibility. It makes me realize what a silo I am in when I never heard of any of these before checking the rising feed, and I consider that a blessing. To be frank, most of the investing and business portion of SS is slop, and this should not surprise anyone. There may be a false positive, but the direction is clear.
For the next experiment, I took a category one would hope would not use AI, cultural commentary. Looking at some of the top Substacks for cuture, this seemed a little better. Of the non-podcast Substacks in English, five were human, one was mixed, and one was slop. Not great, but not catastrophic.
Culture:
Selvaggia Lucarelli - Italian, did not check.
Elizabeth Spiridakis - Human (All paid posts with few substantial blurbs)
Diabolical Lies - Podcast
Jessica Troisfontaine - French. Did not check
I took another example of a category that one would hope was not mostly slop, Education. It was a disaster, with the top five being one human, three slop, and one mixed.
Education:
Ruben Hassid - 100% Human (Ironic, given this is on how to use AI)
In short, Substack is bursting at the seams with AI.
These are interesting metrics, but what really mattered to me is what people in my actual feed are doing, the ones I have conversations with and supposedly align in the same cultural and political sphere. I ran with a “benefit of the doubt” view whenever I saw something that smelled not quite right, and unfortunately every “smell” proved correct. There is a ton of AI slop in my feed. We are not talking 10 percent AI or a little bit of assistance, which I can tolerate, but the entire article being written by Claude with basically no human input. If I wanted to talk to Claude, I would talk to Claude. It feels like a betrayal, like a body snatcher taking the place of a brother in arms. I thought a human being was on the other side, but only got messages in ones and zeroes.
Luckily, none of them were in the “top tier” of this space, with people like Dave Greene, John Carter, Librarian of Celaeno, William M Briggs, Morgoth and others being completely free of AI. My biggest disappointment comes with the smaller accounts I wanted to support who ended up being too lazy to write their own stuff, shortchanging both me and themselves. There are also some larger ones in the slop pile. I can handle a little bit of rough editing, and anyone who has read my Substack knows I’m in no position to complain about grammar errors, but when the narrative voice is not even your own, it makes the reader feel sick inside.
I decided not to put these particular people on blast, even though they fucking deserve it. If you are a slop AI poster reading this, you know who you are, and I want to make clear you broke my trust and the trust of your colleagues. And for what? A few likes? A little attention? A few paid subscribers? Why did you sell yourself out for so little?
I know it will just get worse. Anyone on X knows that AI slop is getting out of control, with not even paid subscriptions capable of stemming the tide. It’s almost pointless to read comments now, and Substack will soon be in the same boat. The Human to AI ratio of writing on here will continue to get worse, and more stringent gatekeeping will be necessary to keep the LLMs out.
With every new bit of effort required to keep your feed clean, a few more people will decide it’s not worth it and leave. Not only is it a time sink, but every time one is caught reading an LLM, he is passively inputting machine modes of writing and thinking, and corroding his own writing and thoughts in the process. LLM writing is a body snatcher, and its target is your mind, accessed through unwitting agents you were foolish enough to trust.
I now look with trepidation as Pangram makes its analysis, wondering if I am reading a human mind or a digital one. I wonder who will betray me next. I wonder if the entire sphere will devolve into algorithmic garbage and drive away the good writers forever. When I started on this site a couple years ago, I had a small follower count but a treasure trove of great thinkers I looked forward to reading. I now feel surrounded by human-like body snatchers eating away at my time and my soul. I just wanted to have a meeting of the minds. I just wanted to have a conversation.
Thank you for reading Social Matter
Thanks for including my piece and for taking a look at these AI checkers! Super important given how widely they’re being promoted. Fwiw, my writing is always “100% human” (as anyone familiar with my POV on the subject will tell you) so the "mixed" result is likely a false positive….which is a pretty good illustration of why I'm skeptical of these tools in the first place 😵💫 Appreciate the mention!
I decided to test it with one of mine(I’m still skeptical and am on the free version) and it came back 100% human, so that is a mark in its favor so far.