You can imagine a feedback loop though:

Some people use Gmail auto-replies more > Other people therefore receive emails in the style of these auto-replies > These other people then start manually sending emails according to this style > More people then become happier using auto-reply (since it produces replies in a style they've come to expect)

In this way, an auto-reply feature, even if used by only a subset of the population would start to move language trends.

Maybe there's nothing wrong with this, I don't know. But there is a feedback loop there, which means there's a blurrier line than we'd like between "this AI algorithm does what I'd expect" to "my expectations have become shaped by AI algorithms."