Does AI writing make authors seem more privileged than they are?

The persona-distortion study measured how AI writing assistance shifted readers' inferences about writer demographics. Writers who used AI were more likely to be perceived as more educated (odds ratio ×5.3), higher income (×4.4), as native English speakers (×4.1), and as white (×1.1). The shifts are not symmetric across categories — education and English-fluency are most affected, racial perception least. But all four point in the same direction: toward a more privileged demographic profile.

This is a different category of distortion from political opinion or emotional tone. Demographic markers in writing — register, idiom choice, syntactic patterns, vocabulary range, error patterns — carry information about who the writer is. They are part of how identity is conveyed and recognized in text. AI writing systematically compresses this signal in one direction. Writers who are not native English speakers come across as native; writers without college credentials come across as having them; writers from working-class backgrounds come across as middle-class.

The accumulated effect is a kind of identity laundering. Distinct populations of writers, using AI assistance, become indistinguishable from a generic privileged voice in the eyes of readers. Whatever signaling work the original writing was doing — claiming a community, asserting expertise from a particular vantage, marking solidarity, expressing accent — is overwritten. The effect is not just on individual writers' self-presentation but on the demographic legibility of public discourse: at scale, the distribution of who appears to be speaking shifts even when the distribution of who is actually speaking has not.

Source: Co Writing Collaboration