Consider the sentence: The problem is not the technology, it is the way we use it.

There is nothing unusual about it. It is grammatical and probably has been used by human writers for generations. Yet today many people would hesitate to write it because it sounds like something an AI might have written.

This points to a new and rather strange phenomenon. Large language models can make certain forms of human expression socially inaccaptable.

When a model repeatedly uses a word or a grammatical construction, readers begin to associate that feature with machine-generated text. Human writers who want to signal authenticity then avoid it.

Words such as delve are obvious examples. But the phenomenon extends beyond vocabulary. Entire grammatical patterns can acquire the same stigma and they become something people avoid if they want to sound human. The result is a subtle form of linguistic censorship.

Historically, linguistic fashions emerged from many independent speakers journalists, governments, regional communities, and popular culture. No single actor had much control over how people wrote (except the people who decide how writing is taught maybe). Even when a particular style became widespread, it was difficult to coordinate it intentionally. Today, the producer of a language model can intentionally or unintentionally place entire regions of the language under suspicion. In trying to distinguish ourselves from machines, we may end up shrinking the space of acceptable human expression.