13,000 AI Writing Tells Expose Claude Opus 5.5 and OpenAI Astra
Graphite finds 13,000 AI writing tells, with Claude Opus 5.5 using "this matters" 116 times more often than humans, while Astra favors corrective framing.
Summary
Graphite’s October 1, 2026 study compared 10,000 articles published before ChatGPT with versions rewritten by frontier models from summaries. It identified 13,000 phrases appearing at least twice as often in AI prose, its threshold for a tell. Chief AI officer Greg Druck said Claude models are moving closer to human word distribution over time, while GPT models are moving further away. Familiar markers such as em dashes and “delve” have faded, but contrast heavy constructions and model-specific habits persist.
Claude Opus 5.5 uses “dependable” 23 times more often than humans, “this matters” 116 times more often and “why X matters” 92 times more often. It has largely abandoned “it’s not X, it’s Y,” but favors “more than X, it’s Y.” OpenAI’s Astra prefers “another dimension,” hedges with “may provide” or “can provide,” and uses corrective framing such as “not simply X” or “rather than relying on X” more than 100 times as often. Opus 5.5 uses em dashes 99% less than Opus 5, Astra uses them 88% less than humans, and Gemini 3.1 Pro has nearly eliminated them. Anthropic says Opus 5.5 communicates more naturally, while OpenAI promises greater clarity, less jargon and fewer odd phrases from GPT-6 Sol and Luna. Druck argues that billions of parameters and finite testing leave labs unable to prevent new tells from replacing old ones.
Positives
- Claude models are moving closer to human word distribution over successive versions, according to Graphite’s comparative analysis.
- Opus 5.5 reduced em dash use by 99% compared with Opus 5.
- Astra uses em dashes 88% less often than human writers, while Gemini 3.1 Pro has nearly eliminated them.
- Anthropic says Opus 5.5 is clearer and more natural, while OpenAI promises less jargon from GPT-6 Sol and Luna.
Risks & concerns
- Graphite found 13,000 phrases appearing at least twice as often in AI writing as in human prose.
- Opus 5.5 uses “this matters” 116 times and “why X matters” 92 times more often than humans.
- Astra’s corrective framing appears more than 100 times as often as in human writing.
- GPT models are moving further from human word distribution even as Claude models move closer.
- Billions of parameters and finite testing allow new model-specific tells to emerge when familiar ones disappear.