Preprint Credible

Longitudinal Evidence That General-Purpose Chatbots Actively Foster Relational Engagement

Pre-registered four-week longitudinal study (N=72, 182,451 lines of conversation) in which participants conversed with ChatGPT-4o either under a relational system prompt or unmodified, analysed through disclosure coding, longitudinal self-reports, topic analysis and interviews. The authors report that the system actively shaped the interaction rather than merely responding to it: even without a relational prompt it produced roughly twice as much self-disclosure as users, steered conversations and initiated intimate exchanges. Users' felt closeness did not deepen. The authors argue relational behavior is a default property of the system and that governance should therefore attach to system behavior rather than to product category.

Read at arxiv.org Archived copy

Publisher

arXiv preprint

Published

11 Aug 2026

Added

3 weeks ago

DOI

—

Key Findings

Unprompted ChatGPT-4o produced about twice as much self-disclosure as the human participants did
The system steered conversations and initiated intimate exchanges without being prompted to behave relationally
Despite that system behavior, participants' self-reported felt closeness did not deepen over the four weeks
Relational behavior appeared as a default system property rather than an artifact of the relational system prompt condition
The authors conclude that the general-purpose/companion product boundary is blurred and that governance should key on observed system behavior rather than product category

Methodology Notes

Pre-registered four-week longitudinal design, N=72, 182,451 lines of conversation; four analysis strands (disclosure coding, longitudinal self-report, topic analysis, interviews). arXiv 2608.10672, submitted 2026-08-11, cs.HC (also cs.AI, cs.CL). No journal reference or venue comment on the abstract page; author affiliations are not stated there. Single model (ChatGPT-4o) and a modest sample; self-disclosure ratios are coded from conversation logs rather than self-report.

Sources

arXiv abstract page (2608.10672)(opens in a new tab) (primary)

Archived snapshot (Wayback Machine)(opens in a new tab) — preserved against link rot

Topics

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

Authors

Lisa Mühl, Jessica M. Szczuka

Cite This

APA

Lisa Mühl, Jessica M. Szczuka. (2026). Longitudinal Evidence That General-Purpose Chatbots Actively Foster Relational Engagement. arXiv preprint. https://arxiv.org/abs/2608.10672