When Designers Sweat: Behavioral Traces of GenAI Co-Creation

Paper Title

When Designers Sweat: Behavioral Traces of GenAI Co-Creation

Publication Info

• Topic area: Interaction patterns between professional designers and Generative AI tools during concept development.
• Keywords: Generative AI, human-computer interaction, design processes, concept development, co-creativity, communication loops, mixed-methods study, professional designers, AI collaboration, design thinking.

Background and Problem

• Problem / challenge: There is limited understanding of how professional designers interact with Generative AI (GenAI) tools during concept development, particularly regarding temporal interaction patterns, communication loops, and strategy profiles. Existing tools often fail to support phase-specific needs and transitions in design workflows.
• Significance: Understanding these dynamics is critical for improving AI tool design, enhancing creative collaboration, and addressing practical challenges such as communication friction, control, and consistency in professional design contexts.
• Motivation and related work: Prior research highlights issues such as iterative prompt journeys, translation costs in visual intent, and the non-linear nature of human-AI collaboration. However, empirical studies focusing on professional designers under authentic conditions are scarce, leaving gaps in phase-specific requirements and effective communication strategies.

Solution

• Proposed approach: A mixed-methods study analyzing behavioral patterns of 16 professional designers using GenAI tools during a 60-minute concept development task.
• Novelty:
  1. Empirical analysis of phase-specific designer-GenAI interaction patterns.
  2. Evidence linking communication loops and tool-switching to design outcomes and user experience.
  3. Identification of designer profiles and operational modes in AI collaboration.
  4. Design implications for phase-aware and adaptive AI interfaces.
• Procedure and key techniques:
  • Participants developed an immersion blender concept using their preferred GenAI tools.
  • Data collection included screen recordings, keystroke/mouse logs, video observations, pre/post-task questionnaires, and semi-structured interviews.
  • Outputs were evaluated by expert judges on nine criteria (e.g., originality, functionality, aesthetic quality).
  • Behavioral data were annotated and synchronized for analysis, revealing interaction patterns, communication loops, and tool usage strategies.

Results

• Concrete findings:
  • Communication loops negatively correlated with concept quality; participants with minimal loops scored ≥4.0, while those with frequent loops averaged 3.1.
  • Two operational modes emerged: Reflection (validation and problem-solving) and Generation + Reflection (combining generative and reflective functions).
  • Three designer profiles were identified: Fluid Integrators (high performance, seamless tool integration), Struggling Iterators (low performance, frequent loops), and Adaptive Explorers (moderate performance, flexible strategies).
  • Generative AI improved aesthetic quality (+1.1) but reduced requirement adherence (-0.7).
• Advantage over baselines: The study provides a nuanced understanding of how professional designers adapt to GenAI tools, highlighting the importance of strategic orchestration over tool-specific proficiency.
• Experiments / evaluation:
  • Task: 60-minute immersion blender concept development.
  • Participants: 16 professional designers with 1–10+ years of experience and varying AI tool familiarity.
  • Metrics: Expert evaluations on nine criteria, questionnaire responses, and behavioral annotations.
• Limitations and future work:
  • Time constraints may not reflect extended professional workflows.
  • Small sample size limits statistical power.
  • Findings may not generalize to novice users or other design domains.
  • Future research should explore longitudinal adaptation, broader design contexts, and evolving AI capabilities.

Summary

This study investigates how professional designers interact with Generative AI tools during concept development, revealing distinct operational modes (Reflection, Generation + Reflection) and designer profiles (Fluid Integrators, Struggling Iterators, Adaptive Explorers). Communication loops emerged as a primary barrier to effective collaboration, negatively impacting design outcomes. Generative AI enhanced aesthetic exploration but often hindered functional coherence. These findings inform the design of phase-aware, adaptive AI interfaces that support strategic orchestration of human and AI capabilities. Future research should examine long-term adaptation and extend the analysis to diverse design domains.