Why is Anthropic's public writing style so unlike Claude's?

The post observes that Claude has a distinctive writing style—short punchy sentences and memorable 'eyeball kicks'—and that this style has spread widely, with the author's highlighting tools matching phrase patterns rarely used before Claude but now common. Even models mildly distilled from Claude, like Kimi K3, adopt the voice, while GPT-5 and Gemma 4 write more normally, suggesting it's a semi-deliberate choice by Anthropic.

Yet Anthropic's own public writing—research posts and documents like Claude's Constitution—does not use the Claude voice at all. It's described as articulate, easy-reading prose like a seasoned blogger. The author notes that Claude writes 80% of Anthropic's code, making it curious that the company doesn't use Claude for prose.

The post offers conjectures, from mundane to speculative: Dario (Anthropic's CEO) may prefer his own essayistic voice; using Claude's voice in public statements could hurt brand trust around human control; and there may be an internal constituency influencing Claude's voice for safety and transparency reasons, including concerns about adversarial distillation. The excerpt cuts off before completing this last thought.