- Maeterial or topic. Quantified material, especially business, finance, and sport, read this way most often. Occasionally procedural stories, regional government or legal stories especially. Not typically hyperlocal news (though at times even that), but major-metro or state-level, usually less-significant events.

- Bylines or agency credit. Unfamiliar names, names typical of foreign regions with low wages and a large English-speaking population (Philippines, India, Sri Lanka, etc.). A few agencies specialising in outsourced and/or algorithmically-assisted artical generation are known (see the Wired article below).

- Less word salad than paragraph salad. Suspect articles tend to have consistency within a given sentence, but a larger-scale structure seems missing. Normal journalism has this problem frequently enough that this isn't a certain tell, but it does have a smell about it. Contrast with well-strutcured narrative, say, typical of a New Yorker artical. Much algorithmic writing seems to rely heavily on syntactic sugar. I feel often as if someone's reading me a chart or an event timeline, whilst trying to hide the fact. Much GPT-3 content has this nature. I shows no understanding of the content or context.

I'm sincerly hoping your question isn't asked in hopes of improving the deception ...

https://www.wired.com/2012/04/can-an-algorithm-write-a-bette...

It feels like this, but in words.