What "tells" have you noticed for AI-generated text (i.e., typical words/phrases, or other patterns indicative of GenAI)?

For me, it's not particular use of words or phrases.

It's embellishing (i.e adding more words than there needs to be).

Some models will default to using fluff words (transform, supercharge etc) unless you tell them otherwise.

Reality is, if you're good at writing prompts, you can get AI to write like you do.

In short:

- Lazy AI use = generic, long with unnecessary words
- Goog AI use = indistinguishable from a human

This is why writing now needs to include perspective, stories and lived experience. The things where AI can't speak for you.

There is a consistency of tone and cadence that a human doesn't normally achieve. If the prompt was to generate something 'fun' or 'formal' the AI over applies the concept.

Americanised spelling coming from a person who is not American. 

Unless someone has taken the time to create and use AI personas, as others have mentioned the AI-generated text is often dry and lacks the tells you would typically see in human generated responses. Humor, emotion, phraseology just seems off or bland.

I definitely see a lot of emojis, especially when used as bullet points. Speaking of, AI content always includes a ton of bullet point lists.

Regarding social media post, this seems to be another giveaway:

"I did not expect this!

But yet, here I was! 

And what happened next blew me away..."

... a lot of one line/sentence statements with a cliffhanger-esque ending (often followed by a bullet list ;)).

On a related note, many free AI content analysis tools are really bad at discovering AI generated content. And a lot of human-generated content seems to adapt the GenAI styles of writing.