Claude Opus 5.5 Uses 95% Fewer Em Dashes in Shift Toward More Natural AI Writing

Anthropic’s Claude Opus 5.5 appears to be changing how it writes. New analysis shows that the model uses fewer obvious AI writing patterns, shorter sentences, and simpler language than Claude Opus 5.

Claude Opus 5.5 is already considered one of the strongest models for coding, but the latest data suggests Anthropic is also refining its writing style to make responses feel more natural and less recognizable as AI-generated content.

Claude Opus 5.5 shows fewer AI writing patterns

According to Arena, an AI benchmarking tool, Claude Opus 5.5 displays fewer common AI writing patterns. That could make the model less likely to produce content that readers describe as “AI slop.”

The analysis also found that Claude Opus 5.5 uses shorter sentences and simpler language compared with Opus 5.

Arena analyzed answers from Text Arena’s Advanced Reasoning category for August and September 2026. The results showed that 10 of 12 writing characteristics moved in a more natural direction.

Opus 5.5 nearly eliminates em dashes

One of the most noticeable changes is the sharp decline in em dash usage, a punctuation habit often associated with AI-generated writing.

Claude Opus 5 used 15.2 em dashes per 1,000 words. Opus 5.5 uses just 0.8 per 1,000 words, representing a reduction of about 95%.

Semicolon usage also dropped significantly, from 6.10 to 1.64 per 1,000 words.

Claude Opus 5.5 writes shorter sentences but longer responses

The average sentence length fell from 12.14 words with Opus 5 to 10.03 words with Opus 5.5. However, the newer model has one notable tradeoff: it is more verbose overall.

Average response length increased from 453 words to 481 words, making Opus 5.5 the most long-form model in the comparison.

Anthropic is changing how Claude writes

AI models are increasingly focused on coding capabilities, so Anthropic’s apparent effort to change Claude’s writing style is notable. If this trend continues, Claude Opus 5.5 could produce responses that are easier to read and less likely to be identified by familiar AI writing patterns.

Join Mikko Hypponen and security leaders from the NFL, Chanel, and Atlassian for a 2-hour digital summit about what will change with AI speed attacks, what defenders should stop doing, and how to verify, decide, fix, and revalidate at machine speed.

Please reserve your seat