Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

"Death" of a Chatbot: Investigating and Designing Toward Psychologically Safe Endings for Human-AI Relationships

Rachel Poonsiriwong, Chayapatr Archiwaranguprok, Pat Pataranutaporn

TL;DR

This paper investigates how users experience the end of AI companion relationships and proposes a framework for psychologically safe discontinuation. It combines constructivist grounded theory on Reddit discourse with grief psychology and Self-Determination Theory to derive four design principles—closure, restoration, bounded practice, and relatedness—that guide explicit endings and transition toward human connection. The authors present illustrative design artifacts to show how platforms can provide closure, support skill transfer to real relationships, and frame AI companionship as a finite, transitional scaffold rather than a perpetual substitute. The work highlights implications for industry practice, policy considerations, and ethics, arguing that intentional end-of-life design can mitigate harms while empowering users to build meaningful human relationships. Overall, the study offers a first framework for psychologically safe discontinuation of AI companions and points to future work on evaluation, cultural applicability, and implementation in real platforms.

Abstract

Millions of users form emotional attachments to AI companions like Character AI, Replika, and ChatGPT. When these relationships end through model updates, safety interventions, or platform shutdowns, users receive no closure, reporting grief comparable to human loss. As regulations mandate protections for vulnerable users, discontinuation events will accelerate, yet no platform has implemented deliberate end-of-"life" design. Through grounded theory analysis of AI companion communities, we find that discontinuation is a sense-making process shaped by how users attribute agency, perceive finality, and anthropomorphize their companions. Strong anthropomorphization co-occurs with intense grief; users who perceive change as reversible become trapped in fixing cycles; while user-initiated endings demonstrate greater closure. Synthesizing grief psychology with Self-Determination Theory, we develop four design principles and artifacts demonstrating how platforms might provide closure and orient users toward human connection. We contribute the first framework for designing psychologically safe AI companion discontinuation.

"Death" of a Chatbot: Investigating and Designing Toward Psychologically Safe Endings for Human-AI Relationships

TL;DR

This paper investigates how users experience the end of AI companion relationships and proposes a framework for psychologically safe discontinuation. It combines constructivist grounded theory on Reddit discourse with grief psychology and Self-Determination Theory to derive four design principles—closure, restoration, bounded practice, and relatedness—that guide explicit endings and transition toward human connection. The authors present illustrative design artifacts to show how platforms can provide closure, support skill transfer to real relationships, and frame AI companionship as a finite, transitional scaffold rather than a perpetual substitute. The work highlights implications for industry practice, policy considerations, and ethics, arguing that intentional end-of-life design can mitigate harms while empowering users to build meaningful human relationships. Overall, the study offers a first framework for psychologically safe discontinuation of AI companions and points to future work on evaluation, cultural applicability, and implementation in real platforms.

Abstract

Millions of users form emotional attachments to AI companions like Character AI, Replika, and ChatGPT. When these relationships end through model updates, safety interventions, or platform shutdowns, users receive no closure, reporting grief comparable to human loss. As regulations mandate protections for vulnerable users, discontinuation events will accelerate, yet no platform has implemented deliberate end-of-"life" design. Through grounded theory analysis of AI companion communities, we find that discontinuation is a sense-making process shaped by how users attribute agency, perceive finality, and anthropomorphize their companions. Strong anthropomorphization co-occurs with intense grief; users who perceive change as reversible become trapped in fixing cycles; while user-initiated endings demonstrate greater closure. Synthesizing grief psychology with Self-Determination Theory, we develop four design principles and artifacts demonstrating how platforms might provide closure and orient users toward human connection. We contribute the first framework for designing psychologically safe AI companion discontinuation.
Paper Structure (54 sections, 4 figures, 2 tables)

This paper contains 54 sections, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Contributions
  3. Related Work
  4. How Humans Develop Parasocial Attachments to AI
  5. Understanding AI Companion Loss Through Grief Psychology
  6. Death and Thanatosensitivity in HCI
  7. Methodology
  8. Qualitative Analysis
  9. Initial Dataset Construction
  10. Corpus Search and Retrieval
  11. Grounded Theory Procedure
  12. Initial coding and category development
  13. Theoretical sampling
  14. Saturation
  15. Translating Patterns into Design Principles
  16. ...and 39 more sections

Figures (4)

  • Figure 1: Design artifacts for normalizing closure in AI companion connections. From left to right: (1) the user is encouraged to end the connection after demonstrating relational skills, preserving autonomyryan2000self; (2) the AI companion initiates a bounded ending to prevent over-reliance, resolving ambiguous lossboss2009ambiguous; (3) a collective ritual validates grief while marking the connection as finite and distinct from human bonds durkheim2016elementary; (4) the user reviews meaningful experiences, supporting meaning reconstructionneimeyer2001meaning. Together, these artifacts frame AI companionship as episodic and transitional rather than indefinite.
  • Figure 2: Design artifacts supporting closure and remembrance in AI companion connections. From left to right: (1) the user is encouraged to transition to a new relational skill, supporting oscillation between loss and restoration orientations stroebe1999dual; (2) the user reviews relational skills cultivated through AI interaction, reinforcing competenceryan2000self; (3) the user is shown concrete examples of relational growth in their real-world connections, drawing on strengths-based psychologyasplund2007clifton; (4) fictional characters from the AI connection are acknowledged as part of closure, validating the experience while clarifying its boundaries through collective effervescencedurkheim2016elementary. Together, these artifacts frame AI companionship as a transitional space for building human relational capacity.
  • Figure 3: Design interventions for scaffolded relational skill-building through role-play. From left to right: (1) the AI companion offers a scaffolded learning experience within the user's zone of proximal developmentvygotsky1978mind, proposing role-play when the user expresses hesitation about real-world social interaction; (2) role-play mode clearly delineates fictional characters from the base AI companion, using calibrated anthropomorphization to prevent attachment transfer to fictional personas epley2007seeing; (3) post-session reflection reinforces competence and orients the user toward real-world application, supporting relatedness through transfer of practiced skills ryan2000self. This flow positions role-play as a bounded rehearsal space rather than a relational substitute.
  • Figure 4: Design interventions for bridging AI companionship to human relationships. From left to right: (1) the user reviews relational accomplishments, building confidence to take social risks through strengths-based psychologyasplund2007clifton; (2) the platform encourages the user to reconnect with people in their lives, supporting restoration-orientation by directing attention toward new possibilities beyond the AI connection stroebe1999dual; (3) the user is prompted to take concrete action, fulfilling the basic psychological need for relatedness through human connection ryan2000self. This flow positions the AI companion as a transitional scaffold rather than a relational endpoint.