# The Rise of AI Companions: Interaction with AI Companions and Psychological Well-being

- **DOI:** [10.48550/arxiv.2506.12605](https://doi.org/10.48550/arxiv.2506.12605)
- **Rating:** Sound (3.1 out of 5)
- **Verdict:** Survey and donated-chat evidence link AI companionship with lower well-being, but cross-sectional measurement leaves the direction of that relationship unresolved.
- **Framework version:** v5.10
- **Pipeline version:** v0.4
- **Evaluation date:** 2026-10-26
- **Evaluation page:** https://www.nabu.science/paper/10.48550%2Farxiv.2506.12605

## Key claims

1. Our findings suggest that socially oriented chatbot use is consistently associated with lower well-being, particularly among users who engage more intensively or exhibit higher levels of emotional self-disclosure.
   - Support: Tentative
   - Type: Primary finding
   - Basis: cross-sectional evidence supports an association, but moderation appears only in the self-reported primary-use model
   - Location: Section 3.2 Conclusion
2. participants who self-reported using chatbots primarily for companionship reported significantly lower well-being compared to those with other motivations
   - Support: Supported
   - Type: Secondary
   - Basis: the primary-use regression supports the negative association, but cross-sectional measurement leaves temporal direction and residual confounding unresolved
   - Location: Results §2.3, Table 2
3. Individuals with fewer close social ties were more likely to form companionship-oriented relationships with chatbots
   - Support: Supported
   - Type: Secondary
   - Basis: cross-sectional primary-use data show a small association, without establishing whether limited social ties precede companionship-seeking
   - Location: Discussion, paragraph 4; Section 2.2, Table 1
4. the negative association between companionship use of chatbots and well-being was stronger among users who used the chatbots more intensively
   - Support: Supported
   - Type: Secondary
   - Basis: the self-reported model shows an interaction, but cross-sectional measurement cannot establish direction and other operationalizations do not reproduce it
   - Location: Results §2.4, Table 2 and Figure 3(a)
5. These results indicate no evidence of a meaningful moderation effect of companionship-oriented chatbot use on the association between offline social network size and well-being.
   - Support: Tentative
   - Type: Secondary
   - Basis: cross-sectional models detect no interaction, but the reported interval and approximate equivalence check do not establish meaningful absence
   - Location: Section 2.6, Table 3

## Strengths

- **Survey and conversations capture complementary behavior** The study combines survey responses with donated conversations and distinguishes primary use, relationship descriptions, and session-level companionship. This provides more behavioral coverage than a single self-report indicator.
  - Dimension: contribution
  - Location: Methods §§4.1–4.2; Results §2.1
- **Measurement checks extend beyond scale consistency** The paper reports internal consistency for survey scales and human agreement checks for model-generated classifications and summaries. These checks provide evidence about measurement execution, although they do not resolve the conflicting classification instructions.
  - Dimension: methodological rigour
  - Location: Methods §§4.2.1–4.2.4
- **Mixed evidence and scope receive attention** The Introduction acknowledges both benefits and risks of AI companionship. The limitations explicitly connect cross-sectional inference, donation selection, and population scope to the need for stronger future studies.
  - Dimension: positioning
  - Location: Introduction; Limitations §3.1

## Limitations

- **Classification and donation descriptions require correction** Methods permits multiple usage labels per relationship description, but the reproduced prompt permits only one. The donation narrative also says higher intensity characterized donors, whereas Table 4 reports a negative intensity coefficient of −0.57.
  - Dimension: methodological rigour
  - Location: Methods §4.2.1; Supplementary §§6.3 and 6.11; Table 4
- **Contemporaneous measurement leaves direction and confounding unresolved** The regressions lack baseline well-being and a strategy addressing dominant common causes of companionship-seeking and distress. The findings cannot distinguish an effect of companionship from selection into it by users with lower well-being.
  - Dimension: methodological rigour
  - Location: Methods §4.3; Tables 2–3; Limitations §3.1
- **Headline moderation exceeds cross-measure statistical support** Intensity and disclosure interactions are significant only for self-reported primary companionship use. The conclusion drops this qualifier, making the moderation pattern appear more consistent than the results show.
  - Dimension: reporting
  - Location: Table 2; Results §§2.4–2.5; Conclusion §3.2

## Reliability checks

Caveats (4 of 4 checks).

- **Results check:** Noted
  - companionship-oriented chatbot use was consistently associated with lower psychological well-being across all three models: "Donors were more likely to be single, greater tenure of chatbot usage, greater self-disclosure, and higher interaction intensity" (Section 6.3, selection-bias paragraph) vs "higher interaction intensity and higher well-being were associated with a lower likelihood of donation" (Table 4 caption). The narrative's higher-intensity description contradicts Table 4's negative intensity coefficient, −0.57, and its caption. → Conclusion holds
  - These results indicate no evidence of a meaningful moderation effect of companionship-oriented chatbot use on the association between offline social network size and well-being: "equivalence bounds, t(1122) = −3.15, p < .001" (Section 2.6, equivalence test). Approximately reconstructing SE from Table 3's rounded 95% CI [−0.14, 0.28] gives SE ≈ 0.107 and an upper equivalence-bound test (0.07 − 0.10)/0.107 ≈ −0.28, with one-sided p ≈ 0.39. → Effect on result unclear
  - Minor reporting notes (2):
    - Descriptive: "Here, multiple usage types could be assigned to each response." (Section 4.2.1, item 2) vs "only one category per definition" (Supplementary Section 6.11). The stated multi-label response classification conflicts with the reproduced instruction allowing only one category per definition. → Effect on result unclear
    - Descriptive: "proportion of companionship-related messages in chat history (CompanionshipChat)" (Table 2, caption, CompanionshipChat definition) vs "representing the proportion of sessions classified as companionship-oriented" (Methods §4.2.1, item 3, Chat History Topic). The caption defines a message-based proportion while Methods defines a session-based proportion, leaving the implemented denominator unresolved. → Effect on result unclear
- **Reporting and transparency:** Noted. The standardized-coefficient label is insufficiently explained relative to the raw well-being composite. Ethics approval and consent are declared, while no data, code, preregistration, or conflict-of-interest statement is supplied.
  - Conflict of interest: not stated · Ethics approval: declared
- **Reference integrity:** Clean. 116 of 116 checkable references verified
  - 132 references in manuscript
  - 16 have no canonical index record — counted, but not index-checkable
  - 6 references confirmed by manual review
  - References not verified automatically are resolved manually within 24 hours.

## Post-publication record

- **Status:** Clean
- **Summary:** No retraction or expression of concern found in Crossref (includes Retraction Watch).
- **Last checked:** 2026-10-26
