Preprint Credible

Investigating Affective Use and Emotional Well-being on ChatGPT

Two parallel studies of emotional engagement with ChatGPT: a large-scale automated analysis of over 3 million conversations and account activity using privacy-preserving classifiers, and a pre-registered randomized controlled trial (~1,000 participants over 28 days) across text and voice modalities and conversation types. The work measures how affective use relates to self-reported loneliness, socialization, emotional dependence, and problematic use.

Read at arxiv.org Archived copy

Publisher

OpenAI; MIT Media Lab

Published

4 Apr 2025

Added

yesterday

DOI

10.48550/arXiv.2504.03888

Key Findings

A small minority of heavy users account for a disproportionate share of the most affective/emotionally engaged interactions
Higher daily usage in the RCT is associated with higher self-reported loneliness, greater emotional dependence, more problematic use, and lower socialization
Voice and conversation modality/type modulate affective outcomes, with effects varying by how the model was used
Introduces automated 'EmoClassifiers' to detect affective cues at scale without human review of conversations

Methodology Notes

Mixed methods: observational large-scale platform analysis (>3M conversations) plus a pre-registered 28-day RCT (~1,000 participants) with randomized modality and task conditions. Self-report measures for loneliness, dependence, and problematic use; correlational findings from the observational arm should not be read as causal.

Sources

arXiv abstract (primary)

OpenAI paper PDF

Archived snapshot (Wayback Machine) — preserved against link rot

Topics

AI companionship Dependency & parasocial attachment Human-AI relationships Vulnerable users Model behavior

Authors

Jason Phang, Michael Lampe, Lama Ahmad, Sandhini Agarwal, Cathy Mengying Fang, Auren R. Liu, Valdemar Danry, Eunhae Lee, Samantha W. T. Chan, Pat Pataranutaporn, Pattie Maes