IntentFlow: Investigating Fluid Dynamics of Intent Communication in Generative AI

TL;DR

This work identifies four core aspects of intent communication in human–GenAI interaction—articulation, exploration, management, and synchronization—through a systematic literature review of 46 papers. It then operationalizes these insights in IntentFlow, a research probe that integrates all four aspects within a writing-task workflow to study how users articulate evolving intents, explore variations, synchronize with outputs, and manage intent configurations over time. A within-subject user study (N=12) shows that comprehensive, interconnected support reshapes user behavior toward verification-driven refinement, reduces cognitive workload, and improves perceived control and intent–output alignment compared with conventional chat-based interfaces. The findings yield design implications for building generative AI systems that support cyclical, transparent, and reusable intent communication across domains beyond writing, advancing practical, interaction-centered approaches to human–AI collaboration.

Abstract

Generative AI shifts interaction toward intent-based outcome specification, despite user intents being inherently vague, fluid, and evolving. While a growing body of HCI research has proposed diverse interaction techniques to support this process, there is limited understanding of what the key aspects of intent communication are and how they interplay to shape users' workflows. To bridge this gap, we first conduct a systematic literature review of 46 HCI papers and identify four core aspects of intent communication support: intent articulation, exploration, management, and synchronization. To investigate how these aspects interplay in practice, we developed IntentFlow, a research probe that embodies all four aspects for a writing task, and conducted a comparative study (N=12). Our action-level behavioral analysis reveals that comprehensive support enables verification-driven refinement and progressive intent curation, reduces cognitive effort, and improves users' sense of control and understanding of intent-output alignment. We conclude with design implications for building generative AI systems that support intent communication as a dynamic, iterative process.

Figures (15)

• Figure 1: Weighted UpSet plot showing how existing systems combine four intent-support aspects---articulation, exploration, management, and synchronization. Each column represents a unique combination of supported aspects, and bar heights indicate the aggregated frequency of occurrences for each combination.
• Figure 2: Articulation support in IntentFlow. The system decomposes a user’s vague input into a high-level goal and a set of low-level intents, externalized as editable components. Users can (a) edit, (b) delete, or (c) add intents to refine their intent articulation. (d) Intents can also be pinned to support intent management across iterations.
• Figure 3: Exploration support in IntentFlow. Intent dimensions are presented through multiple UI controls, including (a) radio buttons, (b) sliders, and (c) tags, enabling users to explore alternative intent configurations. The figure also shows (d) a preview mechanism for synchronization and (e) a pinning option for intent management.
• Figure 4: Synchronization support in IntentFlow. (a) Hovering over intents and intent dimension values highlights corresponding linked parts of the generated output in green (intents) or blue (intent dimension values). (b) Users can roll back to any prior version, which brings the selected output and its associated intents to the latest position in the workflow. (c) Diff view compares old and new outputs. (d) Pagination allows users to navigate and manage multiple output versions over time, supporting intent management.
• Figure 5: Overall interface of IntentFlow. It is split up into three sections: (A) Chat Panel, (B) Intent Panel, and (C) Output Panel.