This study uncovers the methodological and epistemological intersections inherent between Prompt Engineering techniques and Cognitive Linguistics, in light of Large Language Models (LLMs) as generative tools for linguistic hypotheses. The study stems from an epistemological problematic that attempts to monitor the extent to which engineered prompts can function as "semantic stimuli" capable of eliciting and simulating human cognitive structures—chief among them conceptual metaphors and image schemas established by George Lakoff and Mark Johnson—and employing them to generate robust and scientifically falsifiable cognitive linguistic hypotheses. The research paper posits that formulating input prompts based on specific cognitive frameworks (such as conceptual metaphor, mental spaces, and image schemas) enables intelligent systems to simulate human cognitive structures and generate high-precision explanatory hypotheses that mimic conceptual blending spaces. To implement this, a comparative method integrating cognitive computational linguistics and deep learning models was adopted to examine the mechanisms of transition from technical textual input to cognitive linguistic output. This highlights that "Prompt Engineering" is not merely a mute programming tool, but rather a "guided linguistic structure" that stimulates the latent semantic networks within the automated model. The study concludes by presenting an epistemological model that defines the efficiency and limits of automated generation compared to human cognitive analysis, while providing applied frameworks on how to harness the "prompt" as a generative tool in contemporary linguistic research.