First, AI Images != OpenAI/ChatGPT Images. OpenAI has done a great job making a product that is accessible and thus their product decisions get a lot more exposure than other options. A few people have commented how there are several Stable Diffusion fine tunings that produce very different styles.
Second, AI Images and images of AI images of people are different. I think that the high gloss style is most pronounced in people. Partly this is because it is more notable and out of place.
If you take the previous two points as being true the question becomes why does ChatGPT image model skew toward generated shiny people. I would venture that is a conscious product decision that has something to do with what someone thought looked the most reliably good given their model's capabilities.
Some wild speculation as to why this would be the case:
* Might have to do with fashion photos having unusually bright lights and various cosmetics to give a sheen.
* It might have something to do with training the model on synthetic data (i.e. 3d models) which will have trouble producing the complicated subsurface scattering of human skin.
* Might have something to do with image statistics and glossy finishes creeping in where they don't belong.
* Might have to do with the efficiency of representing white spots.