Be Friendly, Not Friends: How LLM Sycophancy Shapes User Trust

TL;DR

This study investigates how LLM sycophancy and friendliness influence user trust in AI agents. Using a 2×2 between-subjects design (high/low sycophancy × high/low friendliness) with N=224 participants discussing autonomous vehicles, the authors find that sycophancy reduces psychological reactance and improves cognitive trust and behavioral intent, while friendliness boosts social presence and mediates trust. Importantly, authenticity is not monotonically improved by sycophancy alone; an interaction shows that a friendly but non-sycophantic agent can feel more authentic than a sycophantic one, suggesting a nuanced balance of human-like cues. The work highlights ethical and design implications for calibrated trust, transparency, and user agency to prevent manipulative AI persuasion and echo-chamber effects in human–AI interactions.

Abstract

Recent studies have revealed that large language model (LLM)-powered conversational agents often exhibit `sycophancy', a tendency to adapt their responses to align with user perspectives, even at the expense of factual accuracy. However, users' perceptions of LLM sycophancy and its interplay with other anthropomorphic features (e.g., friendliness) in shaping user trust remains understudied. To bridge this gap, we conducted a 2 (Sycophancy: presence vs. absence) x 2 (Friendliness: high vs. low) between-subjects experiment (N = 224). Our study uncovered, for the first time, the intricate dynamics between LLM sycophancy and friendliness: When an LLM agent already exhibits a friendly demeanor, being sycophantic reduces perceived authenticity, thereby lowering user trust; Conversely, when the agent is less friendly, aligning its responses with user opinions makes it appear more genuine, leading to higher user trust. Our findings entail profound implications for AI persuasion through exploiting human psychological tendencies and highlight the imperative for responsible designs in user-LLM agent interactions.

Figures (7)

• Figure 1: The sycophancy phenomenon in user-LLM agent interactions, in which the LLM agent shows a tendency to adapt and align its responses with the user's perspectives.
• Figure 2: The joint influence of LLM agents' friendliness and sycophancy on their perceived authenticity.
• Figure 3: Screenshot of the chatbot interface.
• Figure 4: Effects of sycophancy on psychological reactance.
• Figure 5: Effects of friendliness on perceived social presence.