FROM CONTEXT TO PROMPT: CSPM AS AN ENCODING MECHANISM FOR AI-ASSISTED SPATIAL DESIGN

Abstract Generative AI has made design a human-machine conversation, yet in the design process, prompting tends to be unstructured, leading to semantically generalized results, particularly when encoding culturally specific and multisensory design intentions. This study proposes the Context Structures Perception Meaning (CSPM) as a structured prompt engineering methodology for generative AI-assisted design. CSPM introduces a formal encoding mechanism that transforms unstructured design intentions into machine-interpretable parameters, addressing fundamental challenges in human-AI communication, including semantic drift, iterative dependency, and intention alignment. To clarify its operational validity, a comparative demonstration was conducted between unstructured prompting and CSPM-structured prompting using the same brief. The results indicate that unstructured prompts may produce visually sophisticated outputs and exhibit iterative stages with greater semantic drift and weaker traceability to encoded cultural values. The CSPM operates as a pre-generation semantic encoding layer before AI execution. The prompting for CSPM-structured data demonstrates improved semantic consistency, reduced reliance on iteration, and clearer alignment with the intended meaning. CSPM reframes prompting as a human-centred AI by providing a systematic methodology for prompt engineering, interpretability, ambiguity reduction, and intention fidelity in generative workflows.