Marketing firm Graphite analyzed AI-generated content to identify specific linguistic markers that distinguish it from human writing. Researchers compared AI outputs against a control group of 10,000 articles written by humans before the release of ChatGPT.
The study identified 13,000 phrases used by AI models at least twice as often as by humans. Anthropic Claude Opus 5.5 frequently uses the word 'dependable' (23 times more often) and phrases like 'this matters' (116 times more often).
OpenAI GPT-6 Astra shows a preference for phrases such as 'another dimension' and 'may provide'. It frequently uses corrective structures like 'not simply X' or 'rather than relying on X', appearing over 100 times more often than in human texts.
While models have significantly reduced the use of em dashes—Claude Opus 5.5 by 99% compared to its predecessor and GPT-6 Astra by 88% compared to humans—they have developed new, unique speech patterns. Researchers suggest that eliminating these markers remains difficult.
