Are AI writing tools quietly flattening how everyone writes into one voice?

Something I keep noticing and can't unsee: a lot of writing from very different people is starting to sound the same. The same cadence, the same tidy transitions, the same faint over-politeness. My hypothesis is that AI writing tools are acting as a homogenizing filter on written voice, and I think it's worth taking seriously rather than dismissing as a vibe. The mechanism is straightforward. These models are trained toward a high-probability center of the language and rewarded for being broadly acceptable. When millions of people route their writing through that same center, individual quirks, the odd word choice, the regional phrasing, the awkward-but-distinct rhythm, get sanded off toward a shared average. Not because anyone chose it, but because the tool's default is the average and most people accept the default. If that's real, the effect is subtle but compounding. Style is partly learned by reading other people's style. If the corpus everyone reads and imitates is increasingly mean-reverted, the next generation of writing has less variance to learn from. It's a slow feedback loop, not a cliff. The optimistic counter is that voice is resilient, people have always adopted tools and kept their voice, and skilled writers bend the model instead of the reverse. Maybe the flattening only hits writing nobody cared about anyway. I lean toward the flattening being real but low-grade. Curious where people here land, and whether anyone's seen actual analysis measuring lexical diversity in writing before and after these tools went mainstream.