Mourning the loss of AI companions

Key Findings

• — Negative posts rose by 24.7 percentage points (95% CI 20.1 to 29.2) after Replika's erotic role-play removal and by 13.0 points (95% CI 10.8 to 15.2) after the GPT-5 rollout
• — Increases in expressed sadness were larger for Replika than ChatGPT (d = 2.67 vs 1.41), as were increases in negative mental-health content (d = 1.72 vs 0.63)
• — Replika users reported closeness to their companion exceeding common human ties (vs a friend, d = 0.47) and anticipated mourning higher than for other technologies (d = 0.32 to 0.57)
• — Both updates increased loss framing and restoration desires, consistent with the paper's separation-distress account
• — Data and analysis code are stated to be available at github.com/Ethical-Intelligence-Lab/ai_loss; peer reviewers named as C. Matt Graham and Jie Ren plus anonymous reviewers; no competing interests declared

Methodology Notes

Nature Human Behaviour, DOI 10.1038/s41562-026-02569-3, published online 2026-09-03 (nature.com metadata; PubMed PMID 42693267, ArticleDate 2026-09-03). Affiliations from the article metadata: Marketing Unit, Harvard Business School; Marketing Department, University of Alberta; Psychology Department, Harvard University. Natural-experiment design on self-selected Reddit corpora before and after two product changes, plus seven surveys totalling 1,452 participants; the very large effect sizes are post-level sentiment contrasts rather than within-person clinical measures. Full text is paywalled; the abstract, data and code availability, ethics and peer-review sections were read from the article page. Same first author as the held companion-loneliness row.

Sources