Story perspectives

Study Finds Claude Output Uses Key Phrases Far More

By Drooid · Oct 2, 2026, 6:15 PM UTC · How we work

Summary

Story summary

Graphite, a marketing firm, released a study cataloguing 13,000 AI-generated text “tells.”
The study found “this matters” appears 116 times more often in Claude Opus 5.5 output than human writing.
Claude Opus 5.5 uses “dependable” 23-26 times more frequently than human text.
OpenAI’s Astra model includes “does not establish” 275 times more often than Claude outputs.

1 of 1

Story summary

Graphite, a marketing firm, released a study cataloguing 13,000 AI-generated text “tells.”
The study found “this matters” appears 116 times more often in Claude Opus 5.5 output than human writing.
Claude Opus 5.5 uses “dependable” 23-26 times more frequently than human text.
OpenAI’s Astra model includes “does not establish” 275 times more often than Claude outputs.