Your premise is totally plausible in theory: someone can take an AI article and rewrite it to add human flavor. In fact, that's how a lot of modern content is made.

However, AI detectors look for the skeleton of the text. If you just change a few words here and there in an AI draft, the underlying sentence structures and logical transitions remain highly predictable, and a good detector will still flag it as 30% to 70% AI.

To get a 0% AI score, the text usually has to be written from scratch by a human brain. The anger, the specific visual layout of a Japanese YouTube ad, the inclusion of a specific Japanese company name ("節電プロ株式会社"), and the idiosyncratic phrasing mean that if AI was used, the human rewrote it so thoroughly that the original AI draft was completely obliterated.

In short, detectors are confident because this text is messy, angry, highly specific, and slightly quirky—four things that standard AI models are programmed not to be.