What jumps out to me is not the em-dash stat. It’s the tradeoff: fewer obvious AI tells, but longer answers. That sounds like a model getting better at sounding natural while also getting more willing to keep talking. Those two things are not the same, and I’d want to test whether the extra verbosity is actually more useful or just more words with cleaner punctuation.
The em-dash drop is funny because it confirms something people have been complaining about for a while: a lot of “AI writing” is really just style leakage. If Arena’s read is right, Anthropic is sanding down those tells hard. Shorter sentences and simpler wording would make Claude feel less mannered, maybe less essay-ish. That seems like a real product choice, not an accident. Someone at Anthropic probably looked at the way people post-model text and said: stop sounding so much like a template.
But I’m not fully convinced the metric tells the whole story. Less semicolon, fewer em dashes, shorter average sentence length — sure. That’s easy to measure. “Better writing” is a messier claim. A model can get rid of punctuation quirks and still produce text that is bland, repetitive, or over-explains itself. In fact, “answers got longer” makes me suspicious that we may just be seeing a style shift toward more cushioning and more hedging. Maybe that’s good for some tasks. Maybe it’s just friendlier filler.
If you build with Claude, the practical question is whether this makes outputs easier to ship without heavy editing. That’s the only part I’d really care about. If Opus 5.5 is less likely to spray em dashes everywhere and less likely to sound like a polished-but-generic LLM, that saves time. If the cost is longer responses, though, I’d want to know whether the model is actually answering more fully or just expanding the same idea into more tokens.
I also think there’s a broader signal here: model providers are starting to treat “how the model writes” as a product surface, not just a side effect. That’s interesting. Maybe even overdue. People don’t just use Claude for coding or reasoning; they use it to draft, explain, and communicate. The writing style matters. And right now, one of the weirdest parts of the LLM ecosystem is that punctuation choices can become a cultural marker.
Reference: Claude Opus 5.5 uses 95% fewer em dashes, but its answers are getting longer
