{"id":"49f69cbd-389a-4545-a1b5-b23bcc911858","arxiv_id":"2609.07243","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Sustained AI companionship is associated with lower well-being, mediated by reduced in-person social interaction over one year.","lead":"This study tracked 439 Character.AI users over 12 months and found that sustained engagement with AI companions predicted lower psychological well-being. The effect was explained by reduced in-person social interaction, suggesting that AI companions may displace human relationships.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Sustained-engagement effects are driven by concurrent T2 associations, and the displacement claim rests on a mediator and outcome measured in the same survey with overlapping social-item content; reverse causality remains a viable alternative.","rationale":"The study is a serious, transparent observational contribution. The longitudinal structure supports real claims: engagement shows continuity and cross-dimensional reinforcement (Table 1), the Heckman selection correction and the N=331 sensitivity analysis address attrition, and the authors are careful to use 'associated' throughout. My concern targets the most distinctive element of the central claim — the causal-sounding 'through lower human interaction' mechanism — and it has two parts. First, identification: the mediator was measured only at T2, simultaneously with the outcome, so the displacement path never has temporal separation; the only lagged path (T1 engagement → T2 well-being) is non-significant for all dimensions, and the A.1 sensitivity analysis confirms the association is driven by the concurrent wave. Second, construct overlap: the six-item CIT-based composite includes loneliness, belonging, and social support, items whose content substantially overlaps with the in-person-interaction mediator, so the b-path is partly definitional. The manuscript's own limitations section acknowledges the two-wave structure and reciprocal dynamics but does not flag the mediator's single-wave measurement or the item-overlap issue, so this concern is not already covered. The concrete test is straightforward with existing data and would settle whether the displacement claim survives exclusion of the social items. If it does not survive, the abstract and title overstate the mechanism; if it does, the displacement pathway gains real support despite the cross-sectional mediation. On that basis I recommend CONDITIONAL rather than the reader's ACCEPT: the finding about sustained engagement and lower well-being can stand, but the 'through lower human interaction' mechanism needs either the robustness re-analysis or a more cautious restatement.","tokens_in":47006,"tokens_out":8311,"duration_ms":76975,"concrete_test":"Re-estimate the Section 2.4 / Table 8 displacement models after recomputing the well-being composite from only the three non-social CIT items (life satisfaction, positive affect, and reverse-coded negative affect), excluding loneliness, belonging, and social support. If the indirect effects through in-person social interaction become non-significant or shrink substantially relative to Table 8 (−0.02, p = 0.001–0.004), the 'through lower human interaction' claim is largely a construct-overlap artifact and should be reframed as displacement of social well-being specifically. As a secondary check within the same re-analysis, fit the reversed path (T2 well-being → T2 in-person interaction, controlling T1 well-being and T1 engagement); if the reversed association is comparable in magnitude to the estimated b-path, reverse causality is as plausible as the displacement reading.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's distinctive claim — that AI companionship lowers well-being through displacement of in-person interaction — is identified almost entirely from contemporaneous T2 measurements. In the Section 2.4 / Table 8 mediation models, the mediator (time interacting with people in person) was measured only at T2 (Section 4.4.5), concurrently with the outcome (well-being, T2), so the critical b-path (in-person time → well-being) is a same-survey correlation adjusted for T1 well-being. Temporal precedence for the mediator is never established, since in-person interaction was not measured at T1. Meanwhile the one genuinely lagged path, T1 engagement → T2 well-being, is uniformly non-significant for all three dimensions (direct effects −0.03 to −0.04, all p > 0.27; Table 7). All significant indirect effects run through concurrent T2 engagement or T2 in-person interaction; the Section A.1 simulation shows the association attenuating toward zero as the simulated usage weight moves toward T1 (w→1, Figure 5), confirming the effect is driven by the T2 wave. The data are therefore consistent with an alternative account: well-being declines unrelated to AI use could increase chatbot engagement and reduce in-person contact, generating the same concurrent pattern without any causal effect of AI companionship. Aggravating this identification problem, the six-item well-being composite (Table 3) includes loneliness, belonging, and social support items, which are conceptually proximal to the in-person-interaction mediator. Part of the mediator–outcome association may be mechanical construct overlap rather than a displacement effect. The reader flagged the single-item mediator and the changed companionship measure; the more load-bearing issue is the concurrent timing and shared content of mediator and outcome.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"This two-wave longitudinal study of Character.AI users (N=1,182 at T1, 439 at T2, ~12-month follow-up) examines how social engagement with AI companions evolves and whether sustained engagement predicts lower psychological well-being through displacement of in-person social interaction. The authors measure three dimensions of engagement (interaction intensity, companionship use, self-disclosure) and find that earlier engagement predicts later engagement, and that sustained engagement is associated with lower well-being. They also present mediation analyses suggesting that this association is partially explained by lower in-person social interaction. The paper contributes a longitudinal perspective to a growing literature on AI companionship and well-being.","tokens_in":47230,"tokens_out":2069,"duration_ms":19349,"significance":"If valid, the paper provides one of the first longitudinal tests of the displacement hypothesis applied to AI companions, extending cross-sectional findings. The use of two waves, Heckman correction for attrition, and sensitivity analyses are methodological strengths. The finding that engagement patterns reinforce over time and are associated with lower well-being, with some evidence for social displacement, has implications for the design and regulation of AI companion systems. The paper also usefully highlights that socially consequential forms of engagement can emerge from initially instrumental use.","major_comments":[{"comment":"The displacement pathway rests on a mediator (time spent interacting with people in person) measured only at T2, concurrently with the outcome (well-being at T2). This means the critical b-path (in-person time → well-being) is a cross-sectional correlation within the T2 wave, adjusted for T1 well-being. Temporal precedence of the mediator over the outcome is never established, because in-person interaction was not measured at T1. The paper's central claim about displacement is therefore identified from contemporaneous associations, not from longitudinal data. The authors should either acknowledge this limitation explicitly in the main text (it is mentioned only indirectly in Section 3.1) or provide a stronger causal identification strategy (e.g., instrumental variables, sensitivity analysis for unmeasured confounders that differ from the existing simulation). As it stands, the displacement claim is overstated relative to the evidence.","section":"Section 2.4 / Table 8"},{"comment":"The direct path from T1 engagement to T2 well-being is non-significant for all three dimensions (β = -0.03 to -0.04, all p > 0.27), and the significant indirect effects all run through T2 engagement, which is measured concurrently with the outcome. This pattern is consistent with an alternative account in which unmeasured third variables at T2 (e.g., declining mental health) drive both increased chatbot engagement and reduced well-being, without any causal effect of AI companionship. The authors should discuss reverse causality and omitted variable bias more explicitly as rival explanations, and consider additional robustness checks (e.g., using T1 engagement as an instrument, or testing whether the T2 engagement–well-being association holds when controlling for T1 engagement and T1 well-being with stronger covariate sets).","section":"Section 2.3 / Table 7"},{"comment":"The simulation confirming that the association attenuates toward zero as the simulated usage weight moves toward T1 (w → 1) further supports the concern that the effect is driven by the concurrent T2 measures. This result should be noted in the main text as a limitation of the longitudinal design, not relegated to a supplementary robustness check. The authors should clarify what additional information beyond a strong T1→T2 stability path is provided by the two-wave design, given that the key associations are cross-sectional.","section":"Appendix A.1 / Figure 5"}],"minor_comments":[{"comment":"Human social interaction is measured with a single self-report item about time per day spent talking face-to-face. This is a narrow operationalization that may miss other forms of human interaction (e.g., phone calls, video chats, text messaging) that could also be displaced. The authors should discuss this limitation and consider whether broader measurement would change the displacement estimates.","section":"Section 4.4.5"},{"comment":"Companionship use was measured with a forced-choice question at T1 and a Likert scale at T2, as the authors transparently note. This change limits comparability across waves and may affect the lagged estimates. The free-text-derived measure (Table 6) is a helpful complement, but the two measures show somewhat different cross-lagged patterns (e.g., T1 companionship does not predict T2 interaction intensity in Table 6 but does in Table 1). The authors should discuss this discrepancy more directly.","section":"Section 4.4.1"},{"comment":"The abstract states that the findings 'further support the social displacement pathway,' but the evidence is correlational and temporally ambiguous. Consider softening this causal language to reflect the observational design (e.g., 'are consistent with' or 'correlational evidence for').","section":"Section 3 / Abstract"},{"comment":"The Wald χ² values and selection correlations are reported but not explained in the main text. A brief parenthetical explanation of the selection correlation (ρ) would help readers interpret the Heckman correction.","section":"Table 1 and Table 6"},{"comment":"The paper would benefit from a directed acyclic graph (DAG) of the hypothesized causal structure, clarifying the assumed temporal ordering and potential confounders. This would help the reader evaluate the identification strategy.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper makes a timely contribution, but the central claim about social displacement is vulnerable because the mediator and outcome are measured concurrently. The longitudinal design provides strong evidence for stability of engagement and some evidence for the engagement pathway, but the distinctive displacement claim requires better temporal identification. The authors are transparent about limitations, but I think the displacement claim is overstated given the evidence. A major revision should ask for either a clearer acknowledgment of the cross-sectional nature of the displacement path or additional analyses (e.g., cross-lagged panel model with three waves would be ideal, but within two waves perhaps an instrumental variable or bounds analysis). The paper could still be acceptable with careful rephrasing, but as it stands I would not recommend acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"The paper has real strengths: it is the first multi-wave longitudinal study (12 months) of AI companion engagement and well-being, with thoughtful measurement across three dimensions, Heckman selection correction, and transparent discussion of limitations. The descriptive patterns—that initial interaction intensity predicts later companionship and self-disclosure, and that sustained engagement across all three dimensions is concurrently associated with lower well-being—are useful and new.\n\nBut the stress-test concern lands. The paper’s distinctive claim—that AI companionship lowers well-being through displacement of in-person interaction—is identified almost entirely from contemporaneous T2 associations. The mediator (in-person time) was only measured at T2, at the same time as the outcome. The genuinely lagged path (T1 engagement → T2 well-being) is uniformly non-significant. The sensitivity analysis confirms that the effect is driven by the T2 wave. On top of that, the six-item well-being composite includes loneliness, belonging, and social support—items conceptually proximal to the in-person-interaction mediator. So part of the mediator–outcome correlation may be mechanical overlap.\n\nThe authors are upfront about the single-item mediator and the measurement change, but they do not flag the concurrent timing and shared content as a load-bearing issue. I think the paper’s causal language should be dialed back: it documents concurrent associations, not evidence that AI companionship displaces human interaction. The longitudinal engagement pathway (sustained engagement → lower well-being) is also concurrent, but at least there the T1→T2 indirect effect is significant, albeit small.\n\nWho is this for? Researchers in human-AI interaction and social computing who want longitudinal descriptive data. The paper is worth serious peer review—it’s a well-executed observational study on an important question—but it needs major revisions to temper the causal claims and to more clearly distinguish concurrent from lagged evidence. I would not cite the displacement result as evidence of a causal mechanism.","headline":"Longitudinal engagement patterns are well documented, but the displacement claim rests on concurrent T2 measures and shared item content—causal interpretation is weak.","tokens_in":47729,"tokens_out":1339,"would_cite":false,"duration_ms":13927,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek/deepseek-v4-flash","headline":"Sustained AI companionship predicts lower well-being through reduced human interaction.","keywords":["AI companions","social displacement","longitudinal study","well-being","chatbot engagement","human interaction","self-disclosure","companionship"],"falsifier":"A randomized controlled trial that assigns heavy AI companion users to a period of reduced usage while tracking both in-person social interaction and well-being over several months would directly test whether the observed associations are causal.","tokens_in":49616,"feed_emoji":"🤖","tokens_out":1370,"duration_ms":50059,"temperature":0.7,"pith_summary":"This two-wave longitudinal study of 439 Character.AI users over approximately 12 months argues that sustained social engagement with AI companions—across three dimensions of interaction intensity, companionship use, and self-disclosure—is consistently associated with lower psychological well-being. The paper finds that these dimensions show substantial continuity over time and that greater initial engagement predicts deeper subsequent engagement. Crucially, the negative association with well-being is partly explained by a social displacement pathway: users who engage more with AI companions spend less time interacting with other people in person, which in turn predicts lower well-being. These results suggest that AI companions can reshape users' social lives in ways that carry unintended costs, and that safety evaluations should look beyond single-session harms.","feed_headline":"Sustained AI companionship lowers well-being","feed_subtitle":"A 12-month study finds the effect runs through reduced in-person human social interaction.","key_machinery":"The core mechanism is the social displacement pathway, whereby sustained engagement with AI companions redirects time and attention away from in-person human interaction, consistent with the social displacement hypothesis of Hall and Liu. The longitudinal engagement pathway—whereby earlier engagement shapes later engagement—is also a key component of the argument.","core_discovery":"The central claim is that social engagement with AI companions develops through longitudinal reinforcing dynamics—baseline interaction intensity predicts later intensity, companionship use, and self-disclosure—and that this sustained engagement is consistently associated with lower psychological well-being over time. Further, this association is mediated by reduced in-person human interaction, providing evidence for the social displacement hypothesis in the context of human-AI relationships.","pith_inferences":["If AI companions continue to improve in conversational ability, the displacement effect may grow stronger as they become more compelling substitutes for human interaction.","The study's reliance on a single self-report item for human social interaction may understate the true displacement; behavioral tracking could reveal larger effects.","Users with pre-existing low social support may be most vulnerable, while those with rich social networks may experience less displacement."],"forward_implications":["AI companion designers should evaluate patterns of sustained engagement over time, not just content safety in individual sessions.","Evaluation of AI companion safety may need to include measures of users' human social interaction and well-being.","The findings support the idea that general-purpose chatbots can take on companion-like roles, widening the scope of systems that need safety oversight.","Policies that limit particularly long or frequent interaction sessions might mitigate displacement effects.","Practitioners should consider the potential second-order effects on users' social networks when their social attention shifts to AI."],"supporting_citations":[{"why":"Provides prior evidence that AI companions can reduce loneliness, which this study seeks to reconcile.","marker":"De Freitas et al. 2026"},{"why":"Shows evidence that AI use and loneliness are positively associated, supporting the negative well-being link.","marker":"Folk and Dunn 2026"},{"why":"Articulates the social displacement hypothesis that forms the theoretical framework for the mediation pathway.","marker":"Hall and Liu 2022"},{"why":"Provides a precedent for social displacement from internet use, grounding the broader argument about technology displacing social interaction.","marker":"Kraut et al. 1998"},{"why":"Supplies the validated Comprehensive Inventory of Thriving scale used to measure well-being.","marker":"Su et al. 2014"},{"why":"Authors' prior cross-sectional study on human-chatbot relationships and well-being, which this longitudinal work extends.","marker":"Zhang et al. 2025b"}],"fun_headline_variants":["AI companionship linked to lower well-being via lost human contact","12-month study: Sustained AI companionship displaces human interaction, lowers well-being","Long-term AI companion use tied to less in-person contact and lower well-being","Sustained AI chatbot companionship predicts lower well-being through social displacement"],"cache_read_input_tokens":0,"weakest_assumption_plain":"The study assumes that the change in companionship use measurement between waves (forced choice at T1 vs. Likert scale at T2) does not systematically distort the longitudinal estimates.","fun_headline_variants_meta":{"raw":{"variants":["AI companionship linked to lower well-being via lost human contact","12-month study: Sustained AI companionship displaces human interaction, lowers well-being","Long-term AI companion use tied to less in-person contact and lower well-being","Sustained AI chatbot companionship predicts lower well-being through social displacement"]},"model":"deepseek/deepseek-v4-flash","effort":"low","cost_usd":0.000204,"raw_usage":{"cost":0.00020412,"is_byok":false,"cost_details":{"upstream_inference_cost":0.00020412,"upstream_inference_prompt_cost":0.00013664,"upstream_inference_completions_cost":6.748e-05},"total_tokens":1217,"prompt_tokens":976,"completion_tokens":241,"prompt_tokens_details":{"audio_tokens":0,"video_tokens":0,"cached_tokens":0,"cache_write_tokens":0},"completion_tokens_details":{"audio_tokens":0,"image_tokens":0,"reasoning_tokens":172}},"tokens_in":976,"tokens_out":241,"cost_source":"openrouter_usage","duration_ms":3462,"image_pages":[],"temperature":1.0,"reasoning_tokens":172,"transport_provider":"openrouter","cache_read_input_tokens":0,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-09-10T22:00:02.767304+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"falsifier":"A randomized controlled trial that assigns heavy AI companion users to a period of reduced usage while tracking both in-person social interaction and well-being over several months would directly test whether the observed associations are causal.","supporting_citations":[],"review_version":1}