Across higher education, generative AI (GenAI) is rapidly gaining importance and is increasingly shaping learning, teaching, and assessment practices (Jin et al., 2025; Kasneci et al., 2023). This development reflects a growing recognition that meaningful engagement with GenAI depends not only on access to the tools, but also on students’ competencies, particularly AI literacy and prompt-related skills that support goal-directed interaction (Federiakin et al., 2024; Hershkovitz et al., 2025; Knoth et al., 2024). Despite this development, empirical research on how students interact with GenAI remains limited. Although research syntheses show that GenAI is applied across diverse instructional contexts, many studies continue to prioritize tool capabilities or researcher-designed prompts rather than examining how students initiate human–AI interaction through the prompts they author themselves (Wang et al., 2025). As a result, there is little empirical insight into prompting as a situated student practice in real tasks.

This gap is consequential because prompt engineering is inherently compositional: effective prompts typically combine multiple elements that jointly shape large language model (LLM) behavior. Recent studies emphasize the importance of analyzing the interaction and co-occurrence of prompting strategies rather than isolated techniques (Schulhoff et al., 2025; White et al., 2024). Moreover, human–computer interaction research shows that non-experts often struggle to design effective prompts, indicating that intuitive LLM interfaces do not remove the need for explicit skill development and instructional scaffolding (Zamfirescu-Pereira et al., 2023).

Against this backdrop, the present study conceptualizes prompt literacy as a teachable component of AI literacy in higher education. Prompt literacy is defined as students’ capacity to translate goals into written instructions that steer an LLM toward an intended outcome by making response requirements explicit (Knoth et al., 2024; Walter, 2024). This conceptualization builds on integrative AI-literacy frameworks that emphasize effective collaboration with AI systems, and positions prompting as a core interaction skill in LLM-mediated tasks (Ng et al., 2021). Accordingly, prompt engineering is increasingly framed as an instructional target shaping how productively students engage with GenAI (Cain, 2024; Lee & Palmer, 2025).

Accordingly, this study has two aims: first, to describe which prompt-engineering techniques students use in an authentic task; and second, to identify distinct prompt types based on how these techniques are combined. These aims are addressed through three research questions:

RQ1: Which prompt-engineering techniques do students employ when formulating a prompt for an authentic planning task?

RQ2: Which latent prompt types emerge from patterns in students’ prompting techniques?

RQ3: How are prompt types associated with learner characteristics and AI literacy?

The theoretical framework treats prompts as observable interaction for LLM behavior. Prompt literacy is operationalized using a classification scheme that categorizes prompts by the prompting techniques they contain, including role specification, contextualization, explicit constraints, output-format requirements, and evaluation cues, drawing on higher-education reviews and the broader prompt-engineering literature (Cain, 2024; Lee & Palmer, 2025; Schulhoff et al., 2025; White et al., 2024).

Using a large sample of students enrolled at German higher education institutions, the study provides baseline empirical evidence on student-authored prompting practices and derives a five prompting types based on latent class analysis conducted in Latent GOLD. These findings inform evidence-based Prompt Literacy development and support more tailored instructional guidance across higher-education contexts.

To address the research questions, the study adopted a quantitative, exploratory design examining how higher-education students formulate prompts for GenAI and how resulting prompting types relate to AI literacy and learner characteristics. Data was collected via an online survey administered in Qualtrics. Within the survey environment, a ChatGPT interface was embedded, and participants were asked to freely enter a single prompt they would use to obtain a vacation-planning task. Vacation planning was used because it is a familiar, low-stakes task that requires little domain knowledge, improving comparability across participants and reducing confounding from disciplinary expertise. No scaffolding, examples, or prior instruction on prompting were provided, allowing students’ natural prompt-writing practices to be observed. After quality checks, the final sample comprised N = 395 students enrolled at German higher education institutions. The sample was predominantly female (62.3%), with ages ranging from 19 to 41 years (M = 24.6, SD = 3.8). Most participants were enrolled in bachelor’s (71.4%) or master’s programs (26.6%) across a broad range of disciplines, including business, social sciences, engineering, and health or education-related fields. In addition to the prompt task, the questionnaire captured demographic variables and self-report measures assessing AI literacy, AI self-efficacy, and digital self-regulation (Pinski & Benlian, 2023). Participants’ experience with generative AI varied substantially: 41.2% reported using such tools at least weekly, while 58.8% indicated infrequent or no regular use. Self-reported confidence in interacting with AI showed moderate to high variability (M = 5.72 on a 7-point scale, SD = 1.41). Prompting techniques were coded in MAXQDA using a predefined, classification scheme (e.g., role specification, contextualization, explicit constraints, output-format requirements, evaluation cues, and stylistic requirements) (RQ1). Coding followed a structured codebook and was conducted by trained coders (κ = .83), with reliability checks performed prior to final coding Statistical analyses were conducted in SPSS and Latent GOLD. SPSS was used for descriptive statistics, scale scoring, and preliminary association tests. Prompt types were identified using a bias-adjusted three-step latent class analysis with a proportional maximum likelihood estimator (Bakk et al., 2013; Vermunt, 2010). In the first step, latent classes are estimated from the response variables. In the second step, individuals are classified. In the third step, associations between class membership and external variables (covariates) are analyzed (Bakk et al., 2013; Vermunt, 2010) (RQ2, RQ3).

Regarding RQ1, descriptive coding indicated that students typically combined multiple prompting techniques rather than relying on isolated strategies, underscoring prompting as a compositional practice and supporting the view of prompts as observable interaction designs that configure LLM behavior. The latent class analysis (RQ2) identified five distinct prompting types that differ primarily in specificity, informational richness, and the degree of guidance provided to the model. Class 1 (Baseline Prompting) reflects an average pattern: prompts show moderate structure and detail without strongly distinctive features. Class 2 (Context-Specific Directive Prompting) comprises clear, concrete, and situation-appropriate instructions; prompts explicitly specify the task and provide relevant contextual information, enabling focused interactions. In contrast, Class 3 (Underspecified Prompting) captures comparatively vague prompts with low informational content; students provide few constraints or contextual cues, resulting in broadly formulated requests with limited precision. A qualitatively different pattern emerges in Class 4 (Exploratory Validation-Oriented Prompting): prompts commonly express an initial direction but also signal uncertainty, emphasizing feedback, confirmation, or validation rather than requesting a clearly defined output. Finally, Class 5 (Role-Based Elaborated Prompting) represents the most elaborate style: prompts contain extensive information, specify constraints and expectations in detail, and frequently assign an explicit role or perspective to the model, reflecting a highly structured interaction strategy. Associations with learner characteristics were generally modest (RQ3), but field of study, age, and AI openness showed significant differences. Education/pedagogy students were more prevalent in Class 3, while medical students were over-represented in Class 4. Overall, the five-class solution provides an empirically grounded typology of student prompting behavior and highlights substantial heterogeneity in how students initiate interaction with GenAI. This heterogeneity suggests that one-size-fits-all prompting support is unlikely to be effective and points to the value of differentiated instructional guidance tailored to distinct prompting profiles.

Bakk, Z., Tekle, F. B., & Vermunt, J. K. (2013). Estimating the association between latent class membership and external variables using bias-adjusted three-step approaches. Sociological Methodology, 43(1), 272–311. https://doi.org/10.1177/0081175012470644 Cain, W. (2024). Prompting change: Exploring prompt engineering in large language model AI and its potential to transform education. TechTrends, 68(1), 47–57. https://doi.org/10.1007/s11528-023-00896-0 Federiakin, D., Molerov, D., Zlatkin-Troitschanskaia, O., & Maur, A. (2024). Prompt engineering as a new 21st century skill. Frontiers in Education, 9, 1366434. Hershkovitz, A., Tabach, M., Reich, Y., Lurie, L., & Cholcman, T. (2025). Framing and evaluating task-centered generative artificial intelligence literacy for higher education students. Systems, 13(7), 518. https://doi.org/10.3390/systems13070518 Jin, Y., Yan, L., Echeverria, V., Gašević, D., & Martinez-Maldonado, R. (2025). Generative AI in higher education: A global perspective of institutional adoption policies and guidelines. Computers and Education: Artificial Intelligence, 8, 100348. https://doi.org/10.1016/j.caeai.2024.100348 Knoth, N., Tolzin, A., Janson, A., & Leimeister, J. M. (2025). Prompt engineering literacy: A skill-based perspective on prompt engineering strategies and interactions with generative AI in higher education. Computers and Education: Artificial Intelligence, 6, 100225. https://doi.org/10.1016/j.caeai.2024.100225 Lee, D., & Palmer, E. (2025). Prompt engineering in higher education: A systematic review to help inform curricula. International Journal of Educational Technology in Higher Education, 22, 7. https://doi.org/10.1186/s41239-025-00503-7 Ng, D. T. K., Leung, J. K. L., Chu, K. W. S., & Qiao, M. S. (2021). AI literacy: Definition, teaching, evaluation and ethical issues. Proceedings of the Association for Information Science and Technology, 58(1), 504–509. https://doi.org/10.1002/pra2.487 Pinski, Marc and Benlian, Alexander, "AI Literacy - Towards Measuring Human Competency in Artificial Intelligence" (2023). Hawaii International Conference on System Sciences 2023 (HICSS-56). 3. Schulhoff, S., et al. (2024). The Prompt Report: A systematic survey of prompting techniques. arXiv. https://arxiv.org/abs/2406.06608 Vermunt, J. K. (2010). Latent class modeling with covariates: Two improved three-step approaches. Political Analysis, 18(4), 450–469. https://doi.org/10.1093/pan/mpq025 Walter, Y. (2024). Embracing the future of Artificial Intelligence in the classroom: The relevance of AI literacy, prompt engineering, and critical thinking in modern education. International Journal of Educational Technology in Higher Education, 21(1), 15. https://doi.org/10.1186/s41239-024-00448-3 White, J., et al. (2023). A prompt pattern catalog to enhance prompt engineering with ChatGPT. arXiv. https://arxiv.org/abs/2302.11382 Zamfirescu-Pereira, J. D., Wong, R. Y., Hartmann, B., & Yang, Q. (2023). Why Johnny can’t prompt: How non-AI experts try (and fail) to design LLM prompts. In Proceedings of the 2023 CHI (CHI ’23) (Article 437, pp. 1–21). Association for Computing Machinery. https://doi.org/10.1145/3544548.3581388