Yes, training data is a plausible answer to your own question there, as well as mine above. And that explanation does not support your claims that AI is writing differently than humans, it only suggests training sets vary.

Repeating your thesis three times in slightly different words was taught in school. Using outline style and headings to make your points clear was taught in school. People have been writing like this for a long time.

If your argument depends on your subjective idea of “good writing”, that may explain why you think AI & blog styles are changing; they are changing. That still doesn’t suggest that LLMs veer from what they see.

All that aside, as other people have mentioned already, whether someone is using AI is irrelevant, and believing you can detect it and accusing people of using AI quickly becoming a lazy trope, and often incorrect to boot.