Why AI Detectors Can't Tell Who Did the Homework

Schools are running homework through tools that claim they can spot AI writing, and those tools are wrong often enough, in both directions, that one result shouldn't end the conversation at home.

They flag real writing as fake

These detectors look for writing that's very even and predictable, and that's also what careful writing looks like from a kid learning English or a kid who was taught to write in short, clear sentences. Honest work gets flagged all the time, and a kid has almost no way to prove they didn't use AI.

They miss AI that's been nudged

Asking ChatGPT to sound more casual, or changing a few sentences by hand, is usually enough to get past a detector. So a clean result doesn't tell you much either, and I wouldn't treat a detector as proof either way.

What actually tells you

What really tells you is a conversation, and no app can do that for you. A kid who wrote it can explain why they made a certain choice and what they almost wrote instead. A kid who had AI write it usually can't, and you'll know pretty fast.

The Sunday Script

Get one script every Sunday

The words for one hard conversation, in your inbox every Sunday, and a short note once a month. You can stop anytime.