What an AI detector is actually measuring

An “AI detector” does not compare your essay to a secret ChatGPT database. Most of them estimate how expected the next word was, and how even the sentence lengths are. Those scores have names — perplexity and burstiness — and they are weak evidence.

Predictable writing looks machine-like

Language models like common word sequences. A detector that uses a similar model will call common sequences “likely AI.” Formal student writing, technical documentation, and anything heavily edited toward a style guide also use common sequences. That is why a careful human can trip a detector and a messy human can slip past it.

Why I do not sell “undetectable”

If a tool promises that your text will pass Turnitin or GPTZero, it is selling a moving target. The detector vendors change their models. A rewrite that “works” in March fails in June. More importantly, the buyer is often trying to hide authorship. I will not help with that pitch.

TextFixer’s AI Content Detector is a pattern hint for editors, not a courtroom test. Use it to see whether a draft still sounds generic. Do not use it as proof that a person cheated, and do not use a rewriter to beat it for a graded assignment.

What to do instead

If you are a teacher, ask for process: outlines, drafts, and sources. If you are a writer, keep those yourself. The provenance of a piece is a paper trail, not a percentage in a sidebar.