Editing by prompt. How professional editors use ChatGPT when revising business texts

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Writing technologies shape writing processes and practices. This was true of word processing, speech recognition, and writing with internet-based sources, each of which changed how writers produced and revised text. Generative AI may have an even more fundamental impact because it can participate directly in producing, evaluating, and transforming text. In this study, we examine how generative AI affects the revision processes of professional editors. Six professional editors revised four Dutch business letters with ChatGPT available alongside Word. Using Inputlog process data, ChatGPT interactions, interim texts, and interviews, we reconstructed when and how editors used the model, what revision work they assigned to it, and how generated material entered subsequent editing. Editors used ChatGPT mainly to generate candidate repairs, from complete rewrites to local reformulations, and sometimes to help identify problems in the source text. They differed, however, in when they turned to the model and in how much revision they had already done themselves. Most evaluation, adaptation, and integration still took place in Word. ChatGPT replaced particular revision activities rather than the editing task as a whole. Editors still had to define problems and goals, judge the model’s output, and decide what to retain in the final text. We discuss what these changes mean for how professional revision is organized and for the role of editorial expertise in AI-supported editing.