Opus 5.5 loves to tell you ‘this matters’ (and other AI writing tells)
Graphite study found 13,000 phrases over twice as common in AI writing; Claude Opus 5.5 says "this matters" 116 times more than humans.
Marketing firm Graphite compared AI rewrites of 10,000 pre-ChatGPT articles against human originals to identify model-specific writing tells. Claude Opus 5.5's top tells include "this matters" (116x more common) and "dependable" (23x), while OpenAI's Astra favors "corrective framing" constructions over 100x more often. Labs have largely eliminated em-dashes (Opus 5.5 down 99%), but the total number of tells is holding steady as old ones are removed and new ones emerge.
- 13,000 phrases at least 2x more common in AI content than human content
- Claude models converging toward human word distribution; GPT models diverging
- Em-dash use cut 99% in Opus 5.5; Gemini 3.1 Pro nearly eliminated it
- Every model version exhibits its own unique set of tells
- Organizations
- GraphiteGreg DruckOpenAI AstraAnthropic
- AI models
- Claude Opus 5.5Gemini 3.1 Pro
From the original article
3 min read at techcrunch.com
Opus 5. 5’s biggest tell is the word “dependable,” which pops up 23 times more often than in human samples.
By Russell Brandom · techcrunch.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.