Prompt Adaptation as a Dynamic Complement in Generative AI Systems

The authors thank Vivian Liu for early contributions to this project. The authors are also grateful to Greg Sun, Ethan Mollick, Nicholas Otis, Solene Delecourt, Rembrand Koning, Daniel Rock, Emma Wiles, Sonia Jaffe, Jake Hofman, and Benjamin Lira Luttges for their feedback. The authors have benefited from seminar and conference feedback at AFE 2025, the NYU-TAU Workshop on AI, MIT CODE, UC Berkeley, Microsoft, and the World Bank. E. Jahani, S. Suri, and D. Holtz led, directed, and oversaw the project; J. Zhang designed and built the experiment apparatus; B. S. Manning led the design of the online experiment flow and Qualtrics survey; J. Zhang led the prompt replay process; M. Alsobay led the analysis of prompt text, with contributions from HY. TuYe and E. Jahani; E. Jahani and HY. TuYe led all other data analysis and engineering, with contributions from J. Zhang, D. Holtz, and M. Alsobay; B. S. Manning, S. Suri, and D. Holtz led the writing of the manuscript; and all authors contributed to designing the research and writing the manuscript and supplementary information. Large language models were used in the drafting and editing of this paper. All ideas herein are those of the authors, who take full responsibility for the contents. All of the “target images” were collected from Unsplash, Reshot, Shopify, Pixabay, or Gratisography; all of these images have licenses for free use for commercial and noncommercial purposes. This study was reviewed by the UC Berkeley Committee for Protection of Human Subjects (Protocol 2023-06-16480).