Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)

Researchers at a marketing firm studied how different AI language models write and found thousands of phrases and sentence patterns that appear far more frequently in AI-generated text than in human writing. Each frontier model has its own distinctive writing habits, such as Claude Opus 5.5's tendency to repeatedly explain why things matter and OpenAI's Astra's preference for "corrective framing" that describes things as "not simply X" or alternatives to existing approaches. While AI labs have successfully eliminated some previously common tells like overusing em-dashes, new distinctive patterns continue to emerge with each model version, suggesting that removing all telltale signs of AI writing may be difficult despite the labs' stated efforts to make models write more naturally.