A word to the wise: Crafting impactful prompts for ChatGPT

The integration of ChatGPT into higher education has grown rapidly, yet research on students’ prompting behaviors remains limited. This study examines learner-generated prompts, focusing on linguistic features, prompt engineering strategies, and content-related problems, as well as their impact on response quality, measured by learner satisfaction. A total of 238 prompts from ten Korean EFL graduate students in English education were collected through reflection logs and analyzed using one-way ANOVA, correlation analysis, network analysis, and K-means clustering. Results indicate that learners predominantly used concise prompts with minimal strategy, with English prompts yielding higher satisfaction than those in their first language. High-satisfaction prompts incorporated clear, interconnected strategies, while low-satisfaction prompts featured fragmented ones. Key strategies associated with high satisfaction included contextualization and specifying text style, tone, and length, while repetition correlated with dissatisfaction. Furthermore, a thematic analysis identified six main problems that hindered response quality: lack of context, unclear feedback, dual requests, grammatical errors, abstract vocabulary, and excessive constraints. These findings highlight the significance of clear, contextually rich, and strategically crafted prompts in optimizing interactions with ChatGPT, along with the need to equip EFL learners with distinct prompting techniques tailored for AI-assisted language learning, separate from traditional human communication.