Replying to @johnwilander
That seems relatively unlikely. The targeting would likely take place asynchronously from the writing; and ML models are hard to target like that on purpose.
Also: this is more generally a threat from advertising changing writing, and it’s no worse than clickbait.