PromptFlow: Supporting Novice Prompt Engineering Through Structured and Visual Scaffolding

Large language models (LLMs) are increasingly used by students, yet many novice users struggle to construct effective prompts because prompt engineering is rarely taught explicitly. As a result, novices often lack guidance on translating their intentions into effective interactions with LLMs. Prior work frames prompt-based interaction as a gulf of envisioning, where users struggle to translate abstract goals into concrete prompt specifications, express intentions in model-interpretable ways, and assess generated outputs. To address this gap, this paper introduces PromptFlow, a node-based prompt-scaffolding system that externalizes prompt structure into modular components and supports explicit specification and comparison of prompt variants. A user study with novice participants examined how such interface-level scaffolding influences prompt-engineering practices. Results suggest increased awareness of prompt components, evidence of strategy transfer when participants returned to a standard chat-based interface, and perceived usefulness of scaffolding features for addressing challenges in conventional chat-based prompting.