It is easy to change the system prompt to make the AI talk with a different voice. It is remarkably hard (at least for Claude, I haven't experimented as much with GPT) to get it to not use so many em-dashes like this essay does.

There's no way. Just the first paragraph alone is enough to convince me; it's too well-written and melodious to be AI, with too much original thought:

Today, the demonic vice of the old is not that they are hard and demanding on the youth — instead they do not demand enough from us, and they cannot quite believe that we have not lived up to the little they have demanded. They think too well of our generation.

Without defending the quality of the rest of the essay, it's a great start. LLMs today could never match it.

Style is the wrong diagnostic. Purple prose and em-dashes can be prompted in or out. The harder question is whether the reasoning was committed or generated. A distinctive voice tells you nothing about whether the person actually worked through the argument or had it produced for them. Which is sort of the point the essay is making about students.

This is such an ignorant trope. The last few places I worked ALL used em-dashes as part of house style and I will continue to use them. It's extremely common (and arguably the LLMs do it because it is extremely common).