31/12/25 - 04:30
In the last month or two (end of 2025), I have noticed an increase in LLM writing. In blog posts, in articles, in videos and their comments. Frustratingly, it is very rarely called out, and even more rarely disclosed up-front.
Obviously this is not a new phenomenon. From the moment chatGPT released to the public in November 2022, 'AI slop' started to invade the web. Spam articles could be churned out faster, bot armies could leave less identifiable comments, and well-meaning individuals felt they could create content that they couldn't before, because of lack of faith in their writing skills, or english not being their first language, or myriad other reasons.
However, it's only in the last month that I've personally experienced this intruding into creative online spaces. This December, for the first time, I came across multiple fanfictions where it strongly appeared the author used an LLM to assist with their writing. They did not disclose this, and when I stopped reading and checked the comments, I could not find anyone complaining - on the contrary, they were engaging with the text in good faith. In fairness, the bar for fanfiction writing can often be pretty low.
I've also come across a few video essays on YouTube, and realised with dissapointment a few minutes in, as the narrator reads cliches and re-phrased points, that they used an LLM in their script-writing process.
I am not against LLMs in principle. They have their uses, and I've experimented a fair bit with using them for writing code, researching things, and just generally testing them. However, the only consistent use-cases I find myself returning to are vibe-coding simple scripts and programs, or as a sounding wall to bounce ideas off of, to figure out where to start learning something for myself. What I have found LLMs terrible for, is writing anything creative. Stories, articles, anything past a wikipedia-level explanation - the results are just not very good.
When I realise, a few sentences into an article, or a few moments into a video, that the media I'm consuming is not a fellow human beings thoughts and feelings on a topic, but soulless text from a machine, I feel tricked. I want to hear what you, the creator, want to communicate. I don't care what the LLM outputs in response to your prompt. If you have an idea, and rather than bringing that idea to life, you feed it into an LLM, I would much rather just read the prompt you gave the LLM. At least that was written by you, and offers some insight into your thoughts.
I cannot engage seriously with LLM output, because there is nobody to engage with. There is nobody behind the screen, no ghost in the machine. When you get chatGPT to polish your blog post or video script, and I read it, and attempt to build a mental model of the creator, I don't get the little piece of you, I just get an LLM. I already have a well-rounded mental model of chatGPT and Mistral and the general zeitgeist of LLMs, because I've used them before. I can guess roughly how an LLM will respond to an input, because I have explored a chunk of the latent space of their responses, and they are just not that creative at the end of the day. I want to read real writing from real people, because people are infinitely more varied and complex. Word choices, phrasing, grammer, punctuation - it all offers some insight, some flavour and context, beyond the raw meaning of the words themselves.
Writing is important for the writer too. When your process is idea > LLM, you knee-cap the exploration and development of your idea. It's strangled in the crib, never nurtured and allowed to grow into something more. The LLM packages it up into a cookie-cutter, corporate-approved, sanitised glob of AI slop. When you write for yourself, you must think. You must choose words, construct sentences, organise and build your ideas. That process is part of the reason I'm writing this post - I wanted to give form to the vague dislike of LLM writing that's been rattling around in my head. As I write, I figure out how to explain and communicate this idea, and in turn I learn about it, I figure out what I'm trying to say.
I'm not happy with my writing. I feel like it's a bit of a mess, I'm over-using fractured, comma-laden sentences, and wondering a bit. But that's the point - my writing tells you something about me and my throught process. And when I read or listen to other people's work, that's an important part of the experience I'm seeking. Poor grammar, bad prose, broken english, typos and misspellings, are all preferable and more meaningful to me than AI slop.
The other reason I'm writing this is I don't see this specific sentiment shared online very much. Comments very rarely call out undisclosed use of LLMs. I think this is important to do, because people don't seem to notice it. They engage genuinely, praising the writing or points made, seemingly unaware that they're praising nobody, enjoying something that was not created with intent and care. It's not the worst thing in the world if they got something out of it, but it feels like the equivelant of watching someone praising the marketing points in the commercial break rather than than the show itself. LLMs are an incredible technology, but they are only a tool, and the tool is not the work. I can appreciate the engineering and science that goes into making a hammer, but I am not so impressed by a piece of wood with nails in it. I am impressed when a person uses the hammer, and wields it with skill or uses it create something.
There is endless good content that people have created, and we only have so many trips around the sun. Do you want to use your time consuming LLM output when there's real art from real people available instead?
To qoute from my favourite game:
Machines make perfect food. But perfection is boring. It's the imperfections that truly make something unique. And so I like to occasionally introduce a human touch. A little imperfection...
- Dogen (Mirror's Edge Catalyst, 2016)
"It's not X, it's Y" - I think this is a pretty well known tell for LLM writing, and clearly a favourite in the RLHF process at many LLM companies. Variations include "Not X, just... Y" and similar negations (didn't, wasn't, etc).
I think this one is particularly over-used, sneaking into writing where a human has removed other tells from an LLM draft, perhaps because it's sounds punchy and deep at first glance. It is very rarely deep, in fact it usually has no calories at all. It produces dumb-sounding statements. A negation like this can be used to clear up an easy-to-make misunderstanding of a point, but LLMs produce hollow, shallow statements. "It's not just cliche, it's boring".
An example from a video essay I started watching - "You're not just playing a game, you're moving through a world that communicates without speaking". What is the point being made here? These are not contradictary or mutually exclusive ideas. It's filler. It hasn't actually negated anything, those are just two seperate statements. The second half is even pretty good, so it's a shame it was mangled like this.
A fanfic I tried to read had a lot of sentences like "Not acting erratic, just... standing", "Not chaotic - deliberate", and "[...] tone was flat. Not mocking. Not dramatic. Just... acknowledging". Blergh. What's the point of negating a statement that was never made? Why would you describe something as "He didn't do X, he actually did Y instead." Just say Y from the outset!