Robot prose is currently terrible

Even the best, most expensive Fall 2026 large language models write badly. The ways in which the writing is sloppy and bad appear, in my experience, to line up with their thinking also being sloppy and bad.

It’s worth reveling for a minute in the fact that we get to live in a world where computers read, write, and seemingly understand natural language. It is not done well; but you are surprised to find it done at all.¹

But people can tell you’re using AI to write

So mostly don’t.

The horrible LinkedIn voice is like spam: you know it when you see it, and it grates on the nerves like spam. It reveals your indifference to my time and lack of respect for my attention and therefore reads like a middle finger: probably appropriate sometimes, but only when you mean it, and only when you know the difference. For those still learning to read and write there’s nothing for it but the fullness of time.

This works in layers, of course. There are a bunch of people out there (probably including my whole management chain at work and half my family) looking at this prose you’re reading right now and thinking the same thing about me: it’s so badly executed that it feels awkward to even say anything about it. I mean, statistically speaking, your writing skills are roughly at the (human) median. Hey, mine too. Your spelling mistakes and non-native grasp of English are completely fine and infinitely preferable to slop, even (especially!) if you are self-conscious about your writing.

But at least try to perceive the difference between robot prose and good human prose, and strive for the latter. And please let all the words be your own (type them yourself), even if you had the AI show you a couple examples along the way.