UMD study shows vocabulary cleanup doesn't humanize AI text, Claude skill built

University of Maryland and Google DeepMind study compared 61,608 human vs AI texts from Claude, GPT, Gemini, DeepSeek, Kimi, finding simple vocabulary changes ineffective. A Reddit user built a Claude skill based on this finding.