Just a short thought dump, ‘real’ article coming up soon-ish. TLDR: you can’t decide what’s AI and what’s not. If you don’t have the time to read all of this, just scroll down to the very bottom and read ‘What Can You Take Away?’ It’s just 2 paragraphs.
94.5%
That’s the percentage of content creators who use AI to create content. Out of every 20 faces you’ll meet on a scrolling streak or video binge, 19 of them actively use AI to generate their work in some form or another—whether it’s creating scripts, generating images, or editing. 78% of writing professionals will utilize AI as a part of their workflow, with those who use AI proficiently having a net 64% higher income than those who refuse to integrate AI into their writing.
In fact, out of the 3 statistics I just gave you, all 3 of them came from GPT-5.2 sending me a list of statistics about AI usage in content creation and writing. Of course, AI is a double-edged sword. It amplifies existing human intelligence. If you aren’t smart enough to handle it properly, it will, instead, make you dumber. Perhaps problematically in this context, AI is a commercial tool, one that’s handed to users on a silver platter, waiting to be used. AI is portrayed to be our future—whether it’s a future of abundance or a future of AI-driven armageddon, its usage is practically ubiquitous.
Those who choose not to use AI but to abuse it have sparked a shockwave of paranoia among the general public. This perhaps dogmatic, pervasive disdain for AI, originating from dystopian science-fiction tropes, has disseminated into public opinion. The ratio of Americans who believe AI will stifle creative thinking as opposed to reinforcing it is now 3.3:1. The ratio of Americans who believe AI will vitiate meaningful human connections as opposed to improving them has now reached 10:1. Even the ratio of Americans who believe AI will deteriorate our ability to solve problems as opposed to those who believe it will reinforce them rests at 1.3:1.
61.22%
That’s the percentage of TOEFL essays, coming from foreign-born students, mistakenly identified as AI-written. Amidst all the distrust and negative framing of AI, AI detectors have squeezed into the new market space, with 68% of K-12 teachers using them during the 2023-2024 school year, up 78% from just 1 year prior. AI detectors, however, are far from ideal—besides suboptimal percentages of false positives on both content generated by AI and humans, they’re extraordinarily biased against foreign students; also seeming to flag content written by neurodivergent students more often. But they’re not necessarily the problem.
There are 3 premises that you have to understand when talking about AI detection.
AI content is always sampled and trained on human-written corpora—its goal is literally to replicate human-generated content, to write in a way that’s palatable to human audiences.
AI will reciprocate linguistic biases in its corpora. The LLM is fed a selective group of information with similar linguistic makeup, making that type of text more prone to AI ‘detection’ false positives.
Due to the frequency of exposure to AI content, humans will subconsciously mimic the way AI writes, which will then be continuously fed back into AI training after some initial latency following the deployment of the LLM. This leads to flagging algorithms being extremely hard to produce—mimicry goes both ways.
With public opinion on AI so stark and nuance so low, too many people have come to the conclusion that human-created ‘problems’ require more human solutions. People now shrug off content they see as AI left and right, deepening their cognitive biases.
They resort to thinking they, as frequent users of AI, can act as the arbiters of originality and human thought, having the ‘intuition’ to differentiate human thoughts from artificial ones.
Truly the pinnacle of human arrogance.
A YouTube video with 1.8 million views encapsulates this type of arrogance extremely well. Evan Edinger explains how he can ‘detect’ AI writing instantly, with the clickbaitish title of the video being “I Can Spot AI Writing Instantly — Here’s How You Can Too”. The amount of false confidence in this video is astounding. Not only is the title blatant clickbait, but the entirety of this video is a tangent about ‘red flags’, pushing a narrative that tells viewers that, if there seem to be enough red flags, it’s going to be AI-generated in some way or another—after all, if I’m too stupid to use AI properly and I can only conceive of using it to generate thoughts for me, other people must be doing the same!
The video provides multiple social media posts—one about photography that actually has a high chance of being AI-written or rephrased from AI, a social media post about Duolingo’s transition into AI, and a Twitter X post from British Prime Minister Keir Starmer. Not only are the 2nd and 3rd ones not AI-generated, but the ‘red flags’ that supposedly help viewers detect AI writing are bogus. Mr. Evan Edinger starts off, 24 seconds in, calling his video a ‘masterclass’ on how to catch AI.
The generic ‘flags’ he brings up are em dashes, the ‘it’s not just x, but y’ sentence structure, and the ‘rule of three’. I’ve been using tons of em dashes in my writing since 5th grade. That was nearly 2 years before I found out about ChatGPT. These are basic logical structures that form the basis of coherent writing. Just because AI is intensively trained to produce coherent writing doesn’t mean that coherent writing is a sign of AI. Then the video goes on an anti-intellectual tangent.
Using commas on iMessage? Yeah, that’s a sign of getting a little help from ChatGPT. Going on tangents when writing? Actually, that’s a sign of originality! Humans obviously always get sidetracked, so if some writing never goes on long tangents of unrelated talking points, it’s actually AI! Anything that repeats/rephrases itself too many times? Yeah, that’s called AI! Using proper grammar on the internet? Oh, of course not, I’m going to say that’s not an obvious red flag since I’m so nuanced, but let’s not even rule that out of the window for now.
Remember, every single one of these points can be dismissed by doing a 10-second Google search on how AI works.
AI content is always sampled and trained on human-written corpora, and the entire point of it is to make content like that of humans.
AI will reciprocate linguistic biases in its corpora. The LLM is fed a selective group of information with similar linguistic makeup, making that type of text more prone to AI ‘detection’ false positives. In this case, coherent and logical writing is the victim.
After a while, humans will mimic AI writing, while AI then continues mimicking them back, making detection impossible at face value.
Real, human-written posts are now going to be denounced by arrogant pseudo-intellectuals as AI-written, because these ideas now seem to be vague, uncontested, yet common knowledge among people. These types of videos set a clear precedent for that type of behavior. Giving users an easy way out, a surefire method to flag AI writing themselves, is not going to be possible—the only thing that results from that is actual, quality thoughts and ideas being dismissed, users going, ‘Oh, that guy used AI to generate his content, therefore he has nothing good to say!’ Using clickbaitish titles, promising that by the end of a 10 or 15-minute video, anyone can tell AI writing right away using their own intellect, even if that’s not the intent of the video, is going to be what viewers take away.
When you say that ‘only one’ red flag isn’t enough to tell if a text is AI-generated, and then go on to frame a regular corporate statement as AI-generated by using 2-3 red flags, the takeaway to users is obvious. This is all paired with the fact that Evan Edinger proceeded to call out one of Keir Starmer’s barely 3-sentence tweets that simply used political language with a mirage of ‘red flags’.
This tweet was obviously not AI-generated. Even then, does the Prime Minister of a G7 country have to open ChatGPT from his phone and type in a prompt just to make AI write a 30-word tweet? Of course, Edinger just says that ‘I’m not 100% sure this tweet is AI-generated’. This is an unhealthy dose of skepticism; this is not adding any nuance to a discussion about how we can rule out some AI-generated work. This is making users pivot and choose to go in the wrong direction. After watching videos like these, people will want to take the path of delusion and ignorance.
Narratives like this capitalize on public concern over AI and the inability of AI detectors to effectively detect if writing is AI-generated. This is a narrative we should strive to avoid, and for good reason.
The other day, one of my friends sent me these screenshots of my latest article over Discord. (go check it out at aalx.substack.com)
Again, pointing out 2 em dashes, 2 ‘it’s not x, it’s y’ sentences, and a triplet. Yes, it’s interesting to point out basic logical structures that you could’ve admired anytime before November of 2022, when ChatGPT came out. I would assume that he didn’t mean to accuse me of using AI—I don’t want to assume too much, but he likely learned a thing or two from content creators promising a silver bullet for AI writing.
Though I think he is much too smart to jump to conclusions as fast as most people when applying ‘red flags’ to obscure, faceless creators.
Staying on the topic of the video, “I Can Spot AI Writing Instantly — Here’s How You Can Too”, Mr. Edinger says 56 seconds in that he hears too many ChatGPT-written sentences on ‘small, faceless YouTube channels’, which is why he oh-so-wisely avoids them. It isn’t a shock to hear that unreliable, small YouTube channels that do not want to put in the effort to write thoughtful scripts will turn to ChatGPT for support. If you know for a fact that a creator is using AI to write their scripts/articles (which means more than finding a handful of ‘red flags’), then you should rightfully avoid them. I’m not going to tell you how to prove that, because I don’t know. I am not here to pretend I know.
Listening to voices who pretend to know is to the detriment of small creators and writers. If someone had bought all of what Evan Edinger said in his video, as well as word from other sources warning people about em dashes and triplets, they would’ve definitely dismissed my work as AI-written. I can say that with an alarming rate of certainty. Many other small, faceless creators would also have their work promptly dismissed as AI. When someone takes 10 seconds of empty thought to dismiss your work as AI, it is one of the worst possible outcomes as a content creator. They believe that it is not your work. That the ideas you spread are not your ideas. That the effort you put into creating content was not your effort. Why should they credit you, support you, for the work they think you haven’t done?
What can you take away?
The one thing I will tell you is this—you will not be able to clearly tell whether or not content is generated by AI simply by reading it yourself. This becomes even more impossible day by day. But again, too many small, faceless creators set a shameless precedent and use ChatGPT as a substitute for their own thoughts. What you should do in this situation is just to consume the same niches of content from small and large creators.
Don’t ask yourself, “Was this AI-written?” Ask yourself, “Was this human-written?” Doing this eliminates your cognitive bias towards affirming false positives. Judge ideas based on merit, not origin. And most importantly, just don’t impulsively support creators. Don’t impulsively like and subscribe. Do it if the content was insightful, if is truly valuable to you, if the ideas propagated will stick with you for the next 10 years. It eliminates so much slop from your feed, and it cuts through so much hassle of determining whether content was generated by AI. You’ll only have to confirm that what you’re watching is human-written.
I do care a lot about this substack and WILL NEVER use ChatGPT to write me anything. If I ever do, and you’ve found more than just em dashes or repetition and whatnot, I highly encourage you to unsubscribe from me and to call me out for it. If you’ve made it this far, you’re a legend. You’ve heard me out over the course of 2,000 words, and I hope those words will carry weight. I appreciate anyone who takes the time out of their day to read my articles. To the person reading this, have a great day.
(I was too lazy to cite properly for this article; have the sources instead if you care a lot about them)
Sources:
https://www.digitalinformationworld.com/2025/11/ai-adoption-climbs-in-writing-world-as.html
https://www.allaboutai.com/resources/ai-statistics/ai-writing/
https://www.researchgate.net/publication/353814736_Detection_Theory_A_User’s_Guide
https://arxiv.org/abs/2506.08872
https://hai.stanford.edu/news/ai-detectors-biased-against-non-native-english-writers
https://www.researchgate.net/publication/353814736_Detection_Theory_A_User’s_Guide
I really agree with and respect your pledge to not use AI in your writing! It's so important to maintain human authenticity.
Well articulated and a bit cheeky. I have struggled with the ‘if you use AI images I won’t read you’ crowds viewpoint most. I do use AI images — thoughtfully, with intentional prompts on a paid account.
I agree with you… if you can’t sense the human underneath the writing, it’s probably not worth your time. But dismissing someone’s ideas, lived experience, or creative value simply because AI touched part of the presentation feels reductive.
Thank you for sharing your perspective.