Notablemodel welfare ethicsOpenAI

Graphite Study Analyzes AI Writing Patterns in Claude Opus 5.5 and Astra

Published
Oct 1, 2026 — 17:50 UTC
Also in this story:Graphite

A Graphite study found that Claude Opus 5.5 uses the phrase "this matters" 116 times more frequently than human writing. The analysis compared AI-generated content from Claude Opus 5.5, Astra, and Gemini 3.1 Pro against a control group of 10,000 human-written articles. Notably, phrases deemed common in AI content appeared at least twice as often as in human samples, with 13,000 phrases identified as such. Greg Druck, Chief AI Officer at Graphite, noted that the term "dependable" appears 23 times more often in Claude Opus 5.5 than in human writing. Furthermore, Opus 5.5 employs em-dashes 99% less frequently than human samples, while Astra shows an 88% reduction in their usage. Druck stated that Claude models are getting closer to human word distribution over time, although he emphasized that these tells are not decreasing. Anthropic claims that Opus 5.5 communicates more naturally than its predecessors, while OpenAI anticipates that upcoming GPT-6 versions will offer more clarity and fewer odd phrases. This study highlights the distinct linguistic patterns of current AI models, which may inform developers in refining their AI writing applications.

Summarised from TechCrunch AI's original report by the Turing Wire Newsdesk. Read the original for the full story.

Source: TechCrunch AI