AI writing – New research from Graphite reveals the specific linguistic patterns and phrases that distinguish AI-generated text from human writing in the era of Claude’s Opus 5.5.
The boundary between human thought and algorithmic output is blurring, yet the digital fingerprints of machine generation remain strikingly persistent. Even as Anthropic claims its latest iteration. Opus 5.5. communicates more naturally than its predecessors. a new analysis suggests the ghost in the machine is still easily spotted by the words it leaves behind.
Researchers at Graphite. a growth agency. set out to map these digital tells by comparing 10. 000 AI-generated articles against an identical volume of human-authored content. The findings suggest that while the intelligence behind the screen is evolving, it has cultivated a new, distinct vocabulary. Opus 5.5 maintains over 2,500 specific linguistic patterns that appear at least twice as often as they do in human writing.
Certain single words act as immediate beacons of machine origin. The word “dependable” appears 23 times more often in AI text than in human prose. “Clearer” shows up at 14 times the human rate. followed closely by “matters” at 13 times. and both “quietly” and “practical” at 11 times the human frequency.
The problem intensifies with two-word combinations, where the AI’s tendency to explain concepts through rigid structures becomes glaring. Phrases like “matters because” appear 146 times more often than in human writing, while “especially valuable” reaches 136 times the rate. “This matters” (116 times), “article explains” (91 times), and “builds trust” (83 times) round out the most frequent offenders.
Perhaps most telling are the three-word phrases that define current AI output. While the researchers noted Opus 5.5 used these combinations hundreds of times. human writers in the study sample did not use them even once: “helps to understand. ” “article explains what. ” “this matters because. ” “taken together these. ” and “this guide walks.”.
The gap between human and machine extends beyond vocabulary into the mechanics of sentence construction. Opus 5.5 exhibits higher lexical diversity, longer words, greater word counts, and a heavier reliance on commas. It also clings to a certain mannered. flowery prose—using flourish over direct language 1.6 times more often than human writers. though this is a 37% improvement over the previous Opus 5 model.
Yet, the evolution of these models is causing a shift in traditional indicators. The once-standard tell of overusing em dashes has vanished; Opus 5.5 uses them at a rate of 0.015 per 1. 000 words. a 99% drop from its predecessor. In a final irony. the presence of an em dash is now more likely to indicate a human author than a bot.
AI writing Opus 5.5 Graphite research human writing artificial intelligence content creation writing patterns