Study: AI Models' Writing Tells Include 'Dependable'

Frontier AI models have recognizable writing habits, and a new study from Graphite puts numbers to them. The company analyzed the writing of leading models and identified 13,000 phrases that appeared at least twice as often in AI-generated content as in human writing.

To establish a baseline, Graphite compared the models against a corpus of 10,000 articles published before ChatGPT. Among the findings, Claude Opus 5.5's clearest tell is the word "dependable," which it used 23 times more often than the human samples. The model also used the phrase "this matters" 116 times more often than human writers.

OpenAI's Astra leaned on corrective framing constructions, using them more than 100 times more often than human writing. Punctuation told a different story: Opus 5.5 used em-dashes 99% less often than Opus 5, Astra used them 88% less than human samples, and Gemini 3.1 Pro has almost eliminated em-dashes entirely.

The findings add detail to how vendors describe their own models. In the Opus 5.5 release, Anthropic said the model "communicates more naturally than prior models" and that early users found its writing clearer and easier to follow. OpenAI said users of the GPT-6 versions of Sol and Luna could "expect to see more clarity, less jargon, [and] fewer odd turns of phrase."

Graphite's chief AI officer Greg Druck told TechCrunch that Claude models are getting closer to the human word distribution over time, while GPT models are getting further away.