From Briefing to Prompt: How Users Translate Textual Design Instructions into Text-to-Image Prompts

Text-to-image systems support creative, educational, and professional work, yet how users translate briefs into effective prompts remains underexplored. This study examines prompt formulation and refinement by 23 participants using ChatGPT with DALL·E 3 under two task configurations: low-detail briefings with multiple reformulations and high-detail briefings with one reformulation. A structured literature review and qualitative analysis of 81 prompts, 81 images, interviews, and observation notes, supported by inter-coder agreement, identified recurring strategies, difficulties expressing visual intentions, and refinement patterns. High-detail/single-reformulation tasks produced semantically denser prompts and closer alignment with intended outcomes; low-detail/multiple-reformulation tasks encouraged exploratory prompting and interpretive variation. Because briefing detail and reformulation opportunities varied together, these findings reflect task-configuration differences, not isolated effects of briefing complexity. Framing prompting as cognitive and creative mediation in human–AI interaction, the study proposes a process model spanning briefing interpretation, semantic extraction, visual attribute selection, prompt structuring, generation, and refinement, requiring further validation.