What Copy-Paste AI Does to Your Sense of Authorship
A pre-registered Scientific Reports study from USC Marshall finds that copy-pasting AI output lowers writers' sense of ownership and meaning, while drafting first and refining with AI later does not. The order of operations is the variable.
TLDR A pre-registered Scientific Reports study from USC Marshall, published in March 2026, finds that copy-pasting AI output lowers writers' sense of ownership, self-efficacy, and meaning at work. Drafting first and bringing AI in to refine afterward shows no significant ownership drop versus not using AI at all. The mode of use is the variable, not the use itself. Today’s hook This week, Microsoft turned on agentic mode in Word, Excel, and PowerPoint for every Copilot license. Accenture rolled the same tool out to 743,000 employees. So the practical question for working adults is no longer whether AI writes with you. It writes with you. The question is what that does to the sense of being the author at the end. What the research shows A new paper in Scientific Reports, published online on March 15, 2026, by Elena Hayoung Lee, Yidan Yin, Nan Jia, and Cheryl Wakslak at USC Marshall, ran a pre-registered experiment with 269 participants and a follow-up survey with 270. Each person did occupation-specific writing tasks under one of three conditions: no AI use, copy-and-paste AI use, or “first human, then AI,” meaning they drafted on their own before bringing AI in to refine. N = 269 participants in the pre-registered Scientific Reports experiment, plus a follow-up survey of 270 more, all running occupation-specific writing tasks across three AI-use conditions The finding is simple and uncomfortable. The copy-and-paste group reported significantly lower psychological ownership of their work than the no-AI group. They also reported less self-efficacy and rated the work as less meaningful. The active-collaboration group, the one that drafted first, looked statistically the same as the no-AI group on ownership. The difference between using AI well and using AI badly, by these measures, was not about the AI. It was about the order of operations. Two corroborators are worth naming. A 2025 paper from the University of Waterloo (Joshi and Vogel, CUI ‘25