Why the Robots Think I’m a Robot
Apparently I write like a machine.
I’m 71.
That’s not an excuse.
It’s the explanation.
In the 1960s I attended a grammar school in Wales, where spelling and grammar were, quite literally, beaten into you. You learned to write correctly because the alternative was a ruler across the knuckles. It’s not a teaching method I’d recommend, but I’ll grant it this much: it worked. Correct English stopped being a choice and became a reflex.
From there I joined the army and trained as a Radio Telegraphist. Accuracy wasn’t a nicety in that job. It was the whole point. A garbled signal isn’t just untidy, it can get someone killed. You learn very quickly to say exactly what you mean, in the clearest possible form, with nothing left to guesswork. That habit doesn’t switch off when you leave the job.
I transferred to the Military Police and eventually became an SIB investigator. Formal, structured, precise report writing was my daily work for years. Every statement had to hold up to scrutiny. Every sentence had to mean one thing and one thing only. You don’t get to be ambiguous when what you’re writing might end up in front of a court martial.
From there I moved to Italy, became an EFL teacher, and later a translator, back when translation meant paper dictionaries and thesauri rather than a search box. Good grammar wasn’t a preference in that trade either. It was the product.
I ended my working life managing exams for a Cambridge English Platinum centre, which is to say I spent years assessing, professionally, whether other people’s English met a formal standard.
So that’s a lifetime of training, across several fields, all pointing in the same direction: write clearly, write correctly, write with structure. No wasted words. No ambiguity. No mess.
And that, it turns out, is exactly what gets you flagged.
AI detectors don’t look for truth or authorship. They can’t. They look for statistical patterns: predictable sentence structure, consistent grammar, formal register, low mess. On their terms, the messier and more erratic the writing, the more human it looks. The cleaner and more disciplined it is, the more machine-like it becomes.
Which means the detector isn’t measuring whether a human wrote something.
It’s measuring whether a human wrote it badly enough.
I didn’t spend a lifetime having accuracy drilled into me by sadistic teachers in gowns, sergeants, investigators and exam boards, only to be told by a piece of software that my writing is too correct to be mine.
A machine with no sense of irony whatsoever has decided that competence is suspicious.
If a detector’s logic is “human writing contains errors, therefore correct writing is suspicious”, then it isn’t detecting AI at all. It’s detecting competence and treating it as a red flag.
That’s not a flaw at the margins.
That is the mistaken premise the whole model stands on.
Somewhere out there is a nineteen-year-old who’s never had a ruler across the knuckles, never had to write a statement that would hold up in a court martial, never had their English checked against a Cambridge standard, and their sloppy first draft sails through clean.
Mine doesn’t.
Because I was taught too well.
We built machines to spot machines, then trained them to mistrust anyone who can spell.
We deserve everything that’s coming to us.