Towards Realistic Artificial User-Guided Usability Test Executions Using GenAI
Abstract
Usability testing in human–computer interaction is effective but time‑consuming. Large language models allow
for natural specifications of simulated user interactions to accelerate early‑stage evaluation. This paper introduces vizron, a system that automates qualitative usability testing on web applications through persona‑guided LLM agents acting as artificial users. Building on the Browser‑Use framework, vizron combines visual, structural, and textual inputs to simulate realistic user navigation and generate think‑aloud feedback. Initial findings reveal a tension between goal‑oriented automation and persona realism, motivating a revised architecture that separates technical execution from persona guidance. Additionally, we argue for artificial users based on dynamic mental models rather than static personas only.
Keywords
usability testing; user experience; generative artificial intelligence; large language model; simulating human users; artificial users