The vulnerable pattern is the product warning
Just under 12% of participants named companionship as their primary motive, yet more than half described the chatbot as a friend, companion, or romantic partner. More than 80% of donated sessions involved emotional or social support, showing how relational use can exceed what people report in a survey label.
The strongest negative association appeared when intense use, smaller offline networks, and companionship motives overlapped. The research does not establish causation, but it identifies the users and behaviors that deserve the most careful product safeguards.
Engagement is not evidence of care
Human disclosure can build mutual trust because both people reveal, respond, and accept responsibility. A chatbot cannot offer the same reciprocity, and a product designed to keep the exchange going may reinforce the very isolation that brought the user there.
Developers should measure whether users reconnect with people, access professional help when needed, and leave the interaction in a healthier state. Usage limits, distress detection, clear reminders, and human referral options should be treated as core product functions rather than optional warnings.
Go to the source
Read the evidence behind this analysis. External links open in a new tab.
Stanford Report — AI companions may worsen loneliness for vulnerable users
