跳转至

文章背景与核心概要

随着人工智能技术的普及，AI辅助日记工具能够根据用户的被动感知行为量身定制提示。然而，用户究竟对哪些行为的提示反应最敏感，目前尚不明确。本文作者通过一项为期八周的研究，分析了369篇日记条目，探讨了AI提示在不同行为类型上的实际效果。

研究发现，行为的响应度在很大程度上受“行为可控性”的影响——具体而言，即该行动是取决于他人，还是完全在个人控制之内。此外，写作风格和阐述质量仅在特定的、依赖上下文的行为类别中对成功产生影响。该研究为理解AI驱动的行为干预边界提供了重要的实证参考，指出了AI日记提示最有可能发挥作用的场景。

Not All Nudges Land: Behavioral Controllability and Elaboration Quality in AI-Supported Journaling

Summary

This research paper investigates the effectiveness of AI-supported journaling tools that tailor prompts based on a user's passively sensed behavior. Analyzing 369 journal entries over an eight-week study, the authors explore which types of behaviors actually respond to AI nudges. The findings reveal that behavioral responsiveness is heavily influenced by behavioral controllability—specifically, whether an action depends on other people versus being entirely within an individual's control. Furthermore, writing style and elaboration quality only impact success within specific, context-dependent behavioral categories.

这篇研究论文探讨了AI辅助日记工具的有效性，这类工具能够根据用户被动感知的行为来定制提示。通过对一项为期八周的研究中的369篇日记条目进行分析，作者探讨了哪些类型的行为对AI的提示真正有响应。研究结果表明，行为响应度受行为可控性的巨大影响——具体来说，即一项行动是取决于他人，还是完全在个人控制范围内。此外，写作风格和阐述质量仅在特定且依赖上下文的行为类别中对成功产生影响。

Document Metadata

Field Details
arXiv ID arXiv:2608.12582 [cs.HC]
Subjects Human-Computer Interaction (cs.HC), Artificial Intelligence (cs.AI)
Authors Nadia Mehjabin, Henry Kautz, Subigya Nepal
Submission Date August 12, 2026
DOI 10.48550/arXiv.2608.12582

文档元数据

字段 详情
arXiv ID arXiv:2608.12582 [cs.HC]
学科分类 人机交互 (cs.HC), 人工智能 (cs.AI)
作者 Nadia Mehjabin, Henry Kautz, Subigya Nepal
提交日期 2026年8月12日
DOI 10.48550/arXiv.2608.12582

Abstract

AI journaling tools can tailor prompts to a person's own sensed behavior, but it is unclear which behaviors respond to them. We analyzed 369 journal entries from an eight-week passive sensing study. An LLM labeled each entry as expressing an intention to change a behavior or not, and we measured follow-through against 26 sensor features with a 3-day before/after comparison.

  • Behavioral Dependency: Responsiveness depended most on whether a behavior involves other people. Behaviors that depend on others improved in only 15% to 22% of cases, while behaviors a person can act on alone improved more often (50% to 63%), though unevenly.
  • Writing and Elaboration: How users wrote mattered less. No single text feature separated improved from unimproved entries; writing carried signal only within specific behaviors, most clearly for text messaging and for longer, more personal intention entries.

Because the sample is small, these are treated as exploratory patterns that point to where AI journaling nudges are most likely to work.

摘要

AI日记工具可以根据个人被感知的行为来定制提示，但目前尚不清楚哪些行为会对这些提示做出反应。我们分析了来自一项为期八周的被动感知研究中的369篇日记条目。大语言模型（LLM）对每个条目进行了标注，以判断其是否表达了改变某种行为的意图，并通过前后3天的对比，对照26个传感器特征衡量了执行情况。

  • 行为依赖性： 响应度主要取决于行为是否涉及其他人。依赖于他人的行为在改善的情况中仅占 15% 至 22%，而个人可以独自采取行动的行为改善频率更高（50% 至 63%），尽管分布并不均匀。
  • 写作与阐述： 用户的写作方式影响较小。没有任何单一的文本特征能将有改善的条目与无改善的条目区分开来；写作仅在特定行为中带有信号意义，在短信发送以及更长、更具个人色彩的意图条目中表现得最为明显。

由于样本量较小，这些被视为探索性模式，指出了AI日记提示最有可能发挥作用的方向。

链接与资源