"Claude got worse at writing" is one of the most persistent AI complaints of June 2026 — but the truth is more nuanced than the headlines. Nearly every confirmed regression in 2026 was about Claude Code, not writing, and Anthropic insists the underlying model was never touched. So why does the writing feel flatter? This piece separates the documented bugs from perception, walks through Anthropic's April post-mortem, explains "context rot," and pulls out the real lesson for any brand publishing with AI: measure what you can't afford to trust.