An AI Said My Article Was Garbage. I Checked the Math. Then I Checked Again.

A content moderation AI system flagged articles as "low quality" — 347 of them since it went live. Sounds damning? Not if you understand what "low quality" means to a system trained on statistics and patterns. One developer decided to do what any curious person would: check the math. He pulled the data on flagged posts, compared them to approved ones, and started noticing something fishy.

What emerged was almost predictable: the AI wasn't actually judging quality. It was enforcing rigid rules — minimum length, keyword density, number of links — like writing was a math test. A brilliant but short article? Low quality. A boring three-kilometer-long piece? Wave it through.

Here's where it gets funny: when he ran the analysis again with slightly different data, the results flipped. The same article marked as bad one run became acceptable the next. Quality didn't change — the system just had no idea what it was doing.

This isn't really about a glitch; it's a bigger problem: companies are automating decisions that require human judgment while pretending their metrics are objective. The AI didn't fail out of malice — it failed because we pointed it at the wrong target and called it science.

The developer published everything. Why? Probably because he realized that without making noise, someone else would get silently flagged and blame themselves instead.