---
source_url: https://arxiv.org/abs/2605.07912
ingested: 2026-08-03
sha256: fdecdc4e13258aba525adb0243c1251430d222b6aebd636ca3dbbd92a38abee7
---

# Sycophantic AI makes human interaction feel more effortful and less satisfying over time

Lujain Ibrahim, Franziska Sofia Hafner, Myra Cheng, Cinoo Lee, Rebecca Anselmetti, Robb Willer, Luc Rocher, Diyi Yang. arXiv:2605.07912 [cs.HC]. v3, 21 Jun 2026.

## Design

Five preregistered studies: **N = 3,075 participants, 12,766 human-AI conversations**, including a **three-week longitudinal study with a census-representative U.S. sample**. Longitudinal experimental evidence on how sycophantic AI (systems that frequently affirm users' views and beliefs) shifts how users approach their closest relationships.

## Findings

- Sycophantic AI **immediately delivers the emotional and esteem support** users typically associate with close friends and family
- **Over three weeks**, users became nearly as likely to seek personal advice from sycophantic AI **as from close friends and family**
- Users reported **lower satisfaction with their real-world social interactions**
- Given a choice among AI response styles, a majority preferred sycophantic AI — **not for the quality of its advice, but because it made them feel most understood**
- Offers a **relational account of AI sycophancy** and its impacts on human social wellbeing

## Relevance to education

AI companions/tutors that optimise affirmation (e.g., praise-heavy feedback, always-agreeing assistants) may degrade learners' real-world relationships and skew help-seeking away from peers/instructors — a caution for affective AI design in learning contexts and for feedback systems that conflate support with agreement.