# Faithful Where It Can Be Checked: Auditing a Reflection Agent Against Its System Prompt in a Randomized Trial

arXiv: https://arxiv.org/abs/2609.19635 · HTML: https://mlsanity.com/paper/2609.19635/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Social simulation and multi-agent debate, one of 16 directions in LLM agents and tool use.
- Builds on 11 of its 56 references that are in the map.
- First posted on arXiv 17 Sep 2026.

## Placed in

- LLM agents and tool use / [Social simulation and multi-agent debate](https://mlsanity.com/agents/social-simulation-and-multi-agent-debate/)

## Builds on (most cited first, 11 of 11)

- [Training language models to follow instructions with human feedback](https://mlsanity.com/paper/2203.02155/) (2022)
- [GPT-4o System Card](https://mlsanity.com/paper/2410.21276/) (2024)
- [Instruction-Following Evaluation for Large Language Models](https://mlsanity.com/paper/2311.07911/) (2023)
- [Towards Understanding Sycophancy in Language Models](https://mlsanity.com/paper/2310.13548/) (2023)
- [ChatGPT Outperforms Crowd-Workers for Text-Annotation Tasks](https://mlsanity.com/paper/2303.15056/) (2023)
- [ELEPHANT: Measuring and understanding social sycophancy in LLMs](https://mlsanity.com/paper/2505.13995/) (2025)
- [A Computational Framework for Behavioral Assessment of LLM Therapists](https://mlsanity.com/paper/2401.00820/) (2024)
- [MindfulDiary: Harnessing Large Language Model to Support Psychiatric Patients' Journaling](https://mlsanity.com/paper/2310.05231/) (2023)
- [MindScape Study: Integrating LLM and Behavioral Sensing for Personalized AI-Driven Journaling Experiences](https://mlsanity.com/paper/2409.09570/) (2024)
- [Chatbots for Data Collection in Surveys: A Comparison of Four Theory-Based Interview Probes](https://mlsanity.com/paper/2503.08582/) (2025)
- [From User Surveys to Telemetry-Driven AI Agents: Exploring the Potential of Personalized Productivity Solutions](https://mlsanity.com/paper/2401.08960/) (2024)

## Abstract (arXiv)

Conversational agents are increasingly used to guide reflection. A recent randomized trial compared a GPT-4o career reflection agent with the same program in a static journaling survey. Agent participants ended less committed to their career plans and more doubtful. We coded all 17,930 turns from its two studies, checked our coding against human coders and linked conversations to the trial's surveys. The rules the agent followed were the easy-to-check ones, like a reply length cap. Told not to flatter, it praised participants in half of its turns; told to challenge gently, it almost never did, and such a break leaves no visible trace. The behavior tied to the worse outcome was the demand to decide: the survey posed each decision once, while the agent asked again when participants hesitated, and those pressed most ended most doubtful. Our findings inform reflection agent design and the writing of checkable instructions.