How Does Turning to AI for Companionship Predict Loneliness and Vice Versa?

Key Findings

• — Increased social chatbot use predicted increased emotional isolation at the next wave (b = 0.051, p = 0.006), and higher emotional isolation predicted increased chatbot use (b = 0.086, p = 0.023).
• — Feeling less socially connected predicted subsequent increases in chatbot use (b = -0.155, p = 0.004), but chatbot use did not significantly predict decreases in social connection (b = -0.013, p = 0.369).
• — The two significant effects held in five of six robustness analyses; the sixth returned p = 0.069 and p = 0.054.
• — Most participants never used a social chatbot in a given wave (73%, 73%, 74% and 70% across waves 1 to 4); 979 of 2,149 participants (46%) completed all four surveys.
• — The authors describe the analyses as exploratory and urge caution in drawing strong conclusions.

Methodology Notes

Prolific Academic sample (50% British, 28% American, 14% Canadian, 8% Australian; mean age 39.99); waves ran November 2023 to February 2024, March to June 2024, July to October 2024 and November 2024 to February 2025. Not preregistered. Materials, data and analysis scripts at osf.io/6xsuz; computational reproducibility confirmed by the journal's STAR team. Funded by SSHRC; authors declare no conflicts. Published online 2026-03-23 in Psychological Science 37(4), pages 276 to 286, CC BY-NC 4.0; the Sage page is blocked from the sweep host, so the record was verified through PubMed (41870975), Crossref and the PsyArXiv version 5 preprint, whose abstract matches the PubMed abstract verbatim. PMC embargo until 2027-03-23.

Sources