谷歌以“generative UI”作为Gemini 3的新核心方向，并在研究中量化其早期性能。为了评估自动化界面生成效果，谷歌通过Upwork雇佣数十位设计师，每个网站设计最高支付130美元，平均制作时间为3至5小时，再将相同指令输入Gemini进行对比。受测用户最终更偏好人工设计，但在44%的案例中认为Gemini界面质量可与人类设计相当。同时，大多数受访者更偏好Gemini的图形化呈现方式，而非传统文本长卷或谷歌搜索返回的首条网站结果，显示界面生成能力在可用性上具统计显著的竞争力。

生成式界面被定位为解决AI回复“千字墙”问题的路径，旨在减少传统聊天式输出的长度与低效率。示例包括根据请求自动生成互动行程图、数学小应用或可点击导航菜单。然而，用户目前仅零星接触到这些界面，谷歌也表明此功能初期仅开放更高阶订阅层级，意味着技术展示与实际普及之间仍存在明显落差。测试提示词也带有引导性，例如旅行示例中暗示用户喜爱Studio Ghibli，从而可能影响模型视觉风格输出。

研究结果强调生成式UI对设计行业的潜在冲击：自动化系统在几乎无人工参与的情况下即可达到近半数场景中与专业设计师可比的质量，但也意味着设计实践中原本由人工确保的统一审美与观点可能被弱化。尽管AI已长期承担重复性设计工作，但当界面生成能力扩展到整体体验层级时，是否牺牲“单一设计视角”成为新的结构性问题。

Google’s “generative UI” defines Gemini 3’s new direction, quantified through early performance studies. To evaluate automated interface generation, Google hired dozens of designers via Upwork, paying up to $130 per website, with an average creation time of three to five hours, and fed identical instructions to Gemini for comparison. Test users preferred human designs overall but found Gemini’s output comparable in 44% of cases. Most respondents also favored Gemini’s visuals over text-heavy answers and even over the top website returned by Google Search, indicating statistically notable usability advantages.

Generative UI is positioned as a remedy for AI’s “wall of text” inefficiency, replacing long-form chat responses with concise interactive layouts. Examples include on-the-fly travel planners, math mini-apps, or click-through menus. Yet users encounter these interfaces only sporadically, and Google has stated the feature is initially limited to higher-tier subscriptions, revealing a gap between demonstration and deployment. Some test prompts were suggestive, such as a Japan-trip query that casually mentioned a love for Studio Ghibli, likely steering the model’s aesthetic choices.

Findings highlight potential disruption to the design field: automated systems can achieve human-comparable quality in nearly half of tested scenarios, raising structural questions about design identity. While AI has long assisted with repetitive design tasks, expanding interface generation to full-experience design risks diminishing the unified artistic viewpoint typically provided by human designers.