Evidence-Based Prompt Design for AI Writing Systems
- Presentation
- AI and New Technologies in TC
- 23. April
- 05:00 PM (CEST) - 05:40 PM (CEST)
- finished
-
Professor Lance Cummings
- University of North Carolina Wilmington
Contents
"Structured prompting is dead." This claim circulates as AI models improve—just talk naturally, they'll figure it out. But what do we mean by "structured prompting"? It's more than adding tags to instructions. Structured prompting transforms ad-hoc AI interactions into systematic content operations by breaking prompts into modular, reusable components—task blocks, context blocks, and content blocks—that can be combined, tested, and scaled across teams.
Through controlled experiments comparing semantic tags, natural structure, JSON, and conversational prompts, I've found that structure isn't about rigid formatting—it's a rhetorical choice shaping how AI collaborates with us. This presentation shares test results and a decision framework for choosing structural approaches based on output goals, scale needs, and variance tolerance. Whether you're building documentation agents or designing repeatable workflows, understanding structure as rhetoric becomes essential as AI systems become central to technical communication.
Takeaways
Gain an evidence-based framework for prompt design decisions and learn how different structural choices shape AI collaboration in documentation and content workflows.
Prior knowledge
Experience creating technical documentation and curiosity about AI integration. Prior prompting experience helpful but not required—we'll cover foundational concepts before diving into testing.