I use AI in my writing, and I'm not going to stop

Substack rolled out a feature that lets you push a button and get a score telling you how much of a post was written with AI. If you run this one through it, the score will be high, and it will be right, because I did use AI to write it. But I used it carefully and deliberately, as a thought partner, and all of the thinking in here is mine.

There's an enormous range in how people are using AI right now, and sentence structure just happens to be the part of it a detector can see. At one end are people taking every shortcut available, publishing work they couldn't defend a single line of. What makes that hollow isn't that a machine wrote the sentences. It's that no human did the thinking. At the other end are people who have built systems that actually understand how they work.

Both of those people get a high score. Somebody scans both pieces, sees the same number twice, and decides they're both worthless. That number doesn't tell you how much thought went into the piece, or how many revisions it went through, or whether the person read every sentence and decided, carefully, that it said what they actually meant.

Within days of the feature launching, I started seeing posts on LinkedIn and Instagram giving away free skills to get your writing past it, plus people selling playbooks for the same thing. So the person who published without thinking can run it through a humanizer and carry on exactly as before, while the person who did the thinking gets flagged. That's the exact opposite of what any of us wanted.

Working out whether a piece of writing has anything in it has always been a human judgment. It's slow and it doesn't scale, which is presumably why we built something we could measure instead and hoped it would stand in for the thing we actually care about.