The reasoning
AI detection by vibes has produced an increasingly strange collection of supposed tells:
Em dashes.
Certain vocabulary.
Smooth prose.
Overly symmetrical composition.
A particular lighting style.
Hands.
No hands.
Too much detail.
Not enough detail.
Some clues can legitimately change the probability that something was AI-generated.
The mistake is turning:
“This resembles patterns I associate with AI”
into:
“Therefore AI made it.”
As generative tools improve, humans imitate AI aesthetics, AI imitates human aesthetics, and hybrid workflows become common, visual intuition becomes an increasingly uncertain classifier.
Example
“This article uses em dashes and numbered sections. ChatGPT obviously wrote it.”
Those characteristics may be compatible with AI generation. They're also compatible with a person who likes em dashes and numbered sections.
