The specifics of the influence of prompt engineering integration on the im-plementation of educational projects: Patterns and comparative assessment

Introduction. Despite the active integration of large language models into education, systemic patterns, comparative effectiveness of prompt engineering integration levels, and sustainability of effects remain understudied. The aim of this research is to identify patterns of prompt engineering influence on educational project management, to conduct a comparative assessment of the instrumental, processual, and cognitive levels of integration, and to reveal the factors ensuring the sustainability of the obtained results. Materials and Methods. The research methodology was built on a three-component framework. Theoretical analysis set the frame for the action research, which involved 117 master’s students divided into control and experimental groups. The statistical block included correlation and regression analysis, ANOVA with Tukey’s post-hoc criterion, Student's t-test, and η² calculation. Qualitative shifts were assessed through content analysis of reflective reports–this made it possible to track not only quantitative but also meaningful changes. Measurements were taken three times: before the experiment began, immediately upon completion of the project cycle, and three months later. Results. The empirical data made it possible to identify three stable patterns. A direct relationship was found between the development of prompt competencies and the completeness of project risk identification (r = 0,73; p < 0,01). At the same time, the dynamics of efficiency turned out to be non-linear. The greatest increase occurs in the range of 45–60 prompts per project, while exceeding this threshold leads to a decline in marginal returns (R² = 0,68; p < 0,01). The functional role of prompts changes depending on the phase of the project life cycle: correlation coefficients range from 0.52 to 0.71, indicating that no single prompt solution exists for all stages. The contribution of integration levels to the total variance is hierarchically structured: the cognitive level accounts for 33% (η² = 0,33), the procedural level for 29% (η² = 0,29), and the instrumental level for 14% (η² = 0,14). All intergroup differences are statistically significant (p < 0,001). The stability of the achieved effects differs quite sharply between the groups. Three months after the withdrawal of AI support, the experimental group retained 89,5% of the indicators, while the control group retained only 65,4% (p < 0,001). Factors ensuring this stability include the following: the institutionalization of template solutions, participants’ reflexive practices, and a verification system based on the ‘human-in-the-loop’ model. According to the data, these are the factors that function as predictors of long-term effect. Conclusions. The data obtained enabled to make three proposals that hold the status of scientific generalizations–concerning the identified patterns of prompt engineering integration, the established hierarchy of its structural components, and the mechanisms for sustaining effects over time. Prompt engineering, in the proposed interpretation, is qualified as a multi-level system comprising instrumental, organizational, and cognitive strata, which fundamentally distinguishes the presented concept from approaches developed in prior works. These theoretical constructs and empirical data provide a foundation for further research in the field of human-machine interaction in the field of educational design.