LLM-Driven UX Micro-Simulations Validated via Proxy Corpora

A novel protocol has been introduced for validating UX micro-simulations driven by LLMs, specifically aimed at early-stage teams that may not have the resources for frequent user studies. This pipeline produces organized feedback on customer experiences, including walkthrough steps, friction points, and micro-survey signals, derived from versioned prompts, personas, tasks, and UI snapshots. Due to the limited availability of public usability datasets with task outcomes, the researchers utilize various public proxy corpora—such as app reviews, support tweets, and open-source software issues—to validate simulated friction themes. The lightweight proxy-validation method incorporates two alignment metrics: top-k Jaccard and distributional weighted-Jaccard (W). The study compares lexical, TF-IDF, and multilingual embedding baselines across six proxy datasets, finding that embedding-based alignment achieves a higher W than lexical baselines on key app-review and support-tweet proxies (e.g., W=0.128 vs 0.000 on Gojek). This research, documented in arXiv paper 2608.13563v1, emphasizes the necessity for decision-oriented signals in early-stage product development, presenting a cost-effective approach for iterative design without extensive user testing.