Researchers at marketing firm Graphite analyzed the writing habits of leading AI models, looking for words and phrases that betray machine authorship. The study, which used a corpus of 10,000 articles published before ChatGPT as a human control group, found 13,000 phrases that were at least twice as common in AI content as in human writing.
Each model shows its own signature. Claude Opus 5.5's biggest tell is the word "dependable," which appears 23 times more often than in human samples. It also favors constructions like "more than an X, it's a Y" and phrases such as "this matters" (116 times more often) and "why X matters" (92 times more often).
OpenAI's Astra model, meanwhile, often describes things as having "another dimension," hedges claims with "may provide" or "can provide," and uses what Graphite calls "corrective framing," such as "not simply X" or "rather than relying on X." These constructions were more than 100 times more common in Astra-generated prose.
Notably, all frontier labs appear to have responded to criticism about overusing em-dashes. Graphite found that Opus 5.5 uses the punctuation mark 99% less often than Opus 5, Astra uses it 88% less than humans, and Gemini 3.1 Pro has nearly eliminated it.
Despite these changes, the overall number of tells is holding steady. "It's not like the tells are decreasing," Graphite's chief AI officer Greg Druck told TechCrunch. "They are managing to remove the most well-known tells, but other ones pop up. And every model version has its own."
The persistence of these patterns is notable given the emphasis on natural writing in recent model releases. Anthropic, for example, touted Opus 5.5's ability to "communicate more naturally than prior models," and OpenAI made similar claims for its latest models. The study suggests that while models can drop specific habits, they develop new ones.