How do university students ask GenAI to revise or proofread a text? A framework for analysing prompting strategies

Datos Bibliográficos

ID21726164
AutoresGraham Burton (0000-0002-8501-9207, Free University of Bozen-Bolzano, autor de correspondencia), Francesca Vitali (0000-0002-7991-1290, Free University of Bozen-Bolzano), Cecilia Lazzeretti (0000-0002-0415-9129, Free University of Bozen-Bolzano), Maria Cristina Gatti (0000-0003-1076-7822, Free University of Bozen-Bolzano)
Año2026
Volumen82
Páginas101706
Fecha de publicación2026-07-01
Peer ReviewedSí
Open AccessSí
TipoARTICLE
RevistaJournal of English for Academic Purposes (JOURNAL)
Identificadores de la revistaISSN: 1475-1585 • E-ISSN: 1878-1497
EditorialElsevier BV (PUBLISHER)
DOI10.1016/j.jeap.2026.101706
OpenAlexW7163034398
IdiomaEN
Referencias citadas25

As the presence of Generative AI (GenAI) becomes ever more widespread in higher education, there is an increasing need to understand how students are actually using it during their studies. This paper focuses on the structure and language of prompts employed by students when using GenAI to revise texts they have written, a common, if not universally accepted, use of the technology. Based on an analysis of a dataset of students' interactions with ChatGPT, we propose a framework for the identification of pragmatic elements contained within students’ prompts to achieve what they perceive as the optimal version of their text. Example elements include requests for revision (e.g., improve the grammar ), instructions related to authorship (e.g., requests not to make additions to a text), and genre/register-related instructions (e.g., make it sound academic ). The framework can be used by researchers and EAP practitioners to analyse prompting strategies and patterns in different contexts. To demonstrate this, we apply the framework to the same dataset to provide a case study of prompting for revision by a group of 70 BA, MA, and PhD students. The results showed a large amount of variability among participants, in terms of both the total number and also type of elements included. Prompts were generally limited in their content: a mean of 3.41 total prompt elements per conversation were produced, with a mean of 1.7714 unique elements. The results suggest a need to raise awareness of different prompting strategies and to explore their potential effects on GenAI output

Ask price · Empirical research · English for academic purposes · Digital Humanities and Scholarship · Discourse Analysis in Language Studies · Writing and Handwriting Education