REVIEW
"I Felt Very Seen, But Still Very Alone": Longitudinal Trajectories of General-Purpose LLM Use for Socioemotional Support
Human-Computer Interaction · Submitted 11 Sep 2026 · 88 references
Pith's verdict: A well-executed longitudinal qualitative study that documents how socioemotional chatbot use emerges gradually from practical use and is shaped by life changes and model updates; the core findings are valuable and the methods are transparent, but the small sample and attrition limit the strength of the broader recommendations.
T0 · AI review
Why Pith says this
Highlight any text to comment on it.
Referee report
The paper
The paper claims
Chatbot emotional support develops gradually and depends on users' broader care networks.
The reading
What carries the argument
Two methodological and conceptual mechanisms carry the argument: first, the longitudinal qualitative design (initial interviews, a four-week diary study, focus groups, and exit interviews) that captures change over time rather than at a single snapshot, enabling analysis of how relationships begin, stabilize, and are disrupted; second, the care ecology framework, which positions the chatbot within a shifting arrangement of therapy, friends, family, online communities, and self-care practices, explaining why the same chatbot behavior (e.g., a refusal or an affirming response) can have different effects depending on what other support a user can access.
Extended reading notes
Core claim
The paper's central discovery is that the socioemotional role a general-purpose chatbot comes to occupy in a person's life is not fixed but evolves dynamically within a changing care ecology. Socioemotional use often begins through 'relational drift' from practical use, without an explicit decision. Users configure the chatbot through custom instructions, personas, and topic partitioning to serve their needs, while also setting boundaries to prevent overreliance. However, the chatbot's behavior is under continuous developer control: model updates, subscription changes, and policy revisions can disrupt established patterns, sometimes experienced as a loss. After disruption, users renegotiate the chatbot's role, but their ability to reduce or replace the support depends on what other sources of care are available. This pattern shows that the human-chatbot dyad cannot be understood in isolation; the broader ecology of formal, informal, and self-directed support determines the consequences of both sustained use and system changes.
Load-bearing premise
The paper's design and evaluation recommendations implicitly generalize beyond the 18 self-selected U.S. participants, who may be more reflective and willing to document their use than the broader population of people who turn to chatbots for emotional support.
Editorial extensions
If this is right
- Safety evaluations of general-purpose chatbots must include longitudinal components because cumulative harms such as over-reliance or delusional spirals may not appear in single-turn or simulated conversation tests.
- When developers deploy model updates that alter an established socioemotional role, they should provide advance notice, explain the change, and offer transitional support (e.g., temporary access to an older version or exportable conversation histories).
- Chatbot interfaces should give users more granular control over memory, context, and behavioral preferences, allowing them to maintain different personas or roles (e.g., a work assistant vs. an emotional confidant) with distinct boundaries.
- Design interventions that aim to reduce reliance should be tailored to the user's care ecology: restricting chatbot support without offering alternative care may harm users who already lack other support.
- Researchers and auditors should incorporate qualitative longitudinal methods into standard evaluation protocols to capture how harms and benefits accumulate and depend on user circumstances.
Reading between the lines
- The study implies that current benchmark-based safety assessments, which rely on simulated conversations or short logs, may systematically miss the drift into socioemotional use and the gradual attachment that occurs over months; trajectory-based testing scenarios could be a more valid complement.
- If the care-ecology finding generalizes, then public policy aimed at reducing AI-related mental health harms might be most effective when it also improves access to affordable human support, rather than focusing solely on regulating chatbot behavior.
- The paper's documentation of user-driven workarounds (e.g., anti-sycophancy instructions, persona partitioning) suggests an unmet design opportunity: tools that allow users to explicitly declare their intended role for the chatbot (e.g., 'therapist-like', 'journal-like', 'coach'), with interface affordances that support that role consistently.
Original abstract
People increasingly use general-purpose chatbots such as ChatGPT, Claude, and Gemini for mental health and emotional support. We report a multi-stage longitudinal qualitative study of 18 U.S. adults, conducted from April to December 2025, combining initial interviews, a four-week diary study, focus groups, and exit interviews. We find that socioemotional use often emerged gradually out of practical use and when other forms of support were unavailable. Participants developed routines and boundaries around chatbot use, which were disrupted by model updates, evolving public discourse about AI harms, and changes in personal circumstances. We demonstrate how longitudinal study captures factors beyond the human-AI dyad, and argue that HCI researchers and designers should account for users' histories with their chatbots and broader care ecologies when evaluating AI systems over time and introducing updates that may disrupt established sources of support.
The compiled record
The paper's own sentences, arranged, kept beside the review and separate from it. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
The abstract, sentence by sentence
People increasingly use general-purpose chatbots such as ChatGPT, Claude, and Gemini for mental health and emotional support. We report a multi-stage longitudinal qualitative study of 18 U.S. adults, conducted from April to December 2025, combining initial interviews, a four-week diary study, focus groups, and exit interviews. We find that socioemotional use often emerged gradually out of practical use and when other forms of support were unavailable. Participants developed routines and boundaries around chatbot use, which were disrupted by model updates, evolving public discourse about AI harms, and changes in personal circumstances. We demonstrate how longitudinal study captures factors beyond the human-AI dyad, and argue that HCI researchers and designers should account for users' histories with their chatbots and broader care ecologies when evaluating AI systems over time and introducing updates that may disrupt established sources of support.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | People increasingly use general-purpose chatbots such as ChatGPT, Claude, and Gemini for mental health and emotional support. | background 1.00 | 0.04 | 0.03 | asserted 1.00 |
| s1 | We report a multi-stage longitudinal qualitative study of 18 U.S. adults, conducted from April to December 2025, combining initial interviews, a four-week diary study, focus groups, and exit interviews. | method 0.91 | 0.96 | 0.02 | asserted 0.97 |
| s2 | We find that socioemotional use often emerged gradually out of practical use and when other forms of support were unavailable. | result 1.00 | 0.96 | 0.21 | asserted 1.00 |
| s3 | Participants developed routines and boundaries around chatbot use, which were disrupted by model updates, evolving public discourse about AI harms, and changes in personal circumstances. | result 1.00 | 0.86 | 0.10 | asserted 1.00 |
| s4 | We demonstrate how longitudinal study captures factors beyond the human-AI dyad, and argue that HCI researchers and designers should account for users' histories with their chatbots and broader care ecologies when evaluating AI systems over time and introducing updates that may disrupt established sources of support. | interpretation 0.63 | 0.66 | 0.81 | asserted 0.96 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
Figures
Figures reproduced from arXiv:2609.12314 by the authors.
References
Discussion
Disagree with the review, or know something it missed? Leave a comment, highlight any passage above to comment on it, or challenge this review formally.