AI companion services that most actively encouraged emotional attachment offered relatively few explicit safeguards in a comparative policy analysis. The conclusion concerns documents, not measured user harm.
What researchers studied
Researcher Yigang Liu examined 49 governance documents from 14 dating applications and AI companion platforms. The material included terms of service, privacy policies and community guidelines.
The analysis used critical discourse methods to examine how platforms describe intimacy, encourage disclosure and allocate responsibility. It distinguished dating services connecting people from systems where the AI itself becomes the companion.
What they found
The study identified two models. Dating applications generally treated emotional information as signals to match people, while AI companion services encouraged longer-term personal disclosure and attachment.
In the author’s interpretation, dating applications often limited responsibility for outcomes between users. Companion services could promote emotional bonding while offering fewer clearly specified protections.
The analysis covered 49 documents, not 49 participants. Its findings are qualitative patterns in governance language; it does not calculate emotional-harm prevalence or prove comparative danger.
What it means
These distinctions matter as companies increasingly market AI tools as social or emotional partners. Terms of service and safety policies influence what protections users can expect.
The findings invite questions about disclosure retention, distress handling and support when interactions become harmful. Clearer safeguards could be evaluated alongside product features.
However, the paper does not test a policy intervention or rank individual services for consumer safety. Its contribution is an interpretive comparison of platform governance.
Important limitation
Document analysis cannot show how consistently companies enforce rules or how users experience them. The sample covers 14 platforms, and policies can change.
The theoretical classification is not a direct experimental measurement of emotional extraction. User interviews, enforcement audits and longitudinal research are needed to assess consequences.
Source Information
Study: Governing intimacy as cultural policy: emotional solicitation, extraction, and abdication in dating app and AI companion platform governance
Authors: Yigang Liu
Journal / Source: Frontiers in Sociology
Published: 13 November 2026







