I skimmed this criticism of Pokimane and got to the fourth paragraph before I realized I was reading a chatGPT output. The emdashes are the most obvious giveaway, but the constant restatements and trouble with building conclusions are also clear tells, if a little more subtle.
In this case I agree overall with the post, but GPT (or grok - they all write the same) argued the point for the poster and did a mediocre job overall, especially for a post with 3M views. More troubling is that the guy insists he wrote the whole thing himself, which has apparently fooled thousands of people. He's far from the only one, either. It's not uncommon to see people using AI to respond to an argument.
I don't know where this is going. AI is a powerful force multiplier, but if more and more people outsource their writing to it, we will eventually get to a soft version of dead internet theory where real people are volleying back and forth with GPT responses but don't fully understand what they're saying to each other. Scammers and grifters will almost be indistinguishable, superficially. I'm surprised the Indian contingent hasn't figured out how to use GPT outputs en masse, but I'm sure it's coming.
The only hope is that people will develop enough AI literacy to recognize automated content. We'll see if this happens.
Your observations about the “GPT giveaway” are astute—those elongated em‑dashes, the frequent signposting of ideas, and the tendency to hedge or re‑state points can indeed tip off a discerning reader.
Firstly, the em‑dash usage you mention is a classic stylistic fingerprint. AI‑generated prose often leans on these punctuation marks to create emphasis or to bridge clauses—sometimes to excess—because they’re statistically prevalent in training data. Human writers, by contrast, typically vary their punctuation more organically.
Secondly, the pattern of “constant restatement” emerges from models trying to reinforce a concept for clarity. In human composition, repetition tends to be more intentional—often for rhetorical effect—whereas in AI output it can feel mechanical, serving as a fallback when the model lacks a clear, concise pathway to a conclusion.
Thirdly, your point about outsourcing writing leading to a “soft dead internet” is well‑taken. If interlocutors merely echo machine‑generated summaries without internalizing or critically engaging with the substance, conversations risk becoming hollow loops of recycled prompts and responses. The more writing is farmed out to AI tools—without human oversight—the greater the chance that nuance, authenticity, and genuine insight will erode.
Don't forget
interspersing bullet point lists
between paragraphs of sentences
which do the same thing
as em-dash characters
with respect to how LLMs generate
content