AI models retain distinctive writing habits despite removing familiar tells
A study by marketing firm Graphite found that leading AI models continue to rely on recognizable words, phrases and sentence patterns, even as they reduce older signals such as em-dashes and “delve.” Researchers identified 13,000 phrases that appeared at least twice as often in AI-generated material as in human writing. Graphite’s chief AI officer Greg Druck said Claude models are moving closer to human word distribution, while GPT models are moving farther away.
The researchers compared 10,000 articles published before ChatGPT with versions rewritten by different models from summaries. For Claude Opus 5.5, the strongest signal was “dependable,” used 23 times more frequently than in human samples. The model also used “this matters” 116 times more often and “why X matters” 92 times more often. Its preference for saying something is “more than an X, it’s a Y” remained noticeable.
OpenAI’s Astra more often used phrases such as “another dimension” and cautious formulations including “may provide” or “can provide.” Corrective framing, such as “not simply X” or “rather than relying on X,” occurred more than 100 times as often as in human writing. Although models have sharply reduced em-dash use, Graphite said the overall number of detectable habits remains broadly stable because new ones emerge with each version.
Written by Sakhanda Wire AI from reporting by TechCrunch. Facts, quotes and full context are in the original article below.