As more and more people lean on chatbots for comfort and guidance, experts have raised alarms over the potential pitfalls: erosion of social skills, unhealthy attachment to digital companions offering synthetic (i.e. fake) empathy, unvetted advice and the risk that vulnerable users might substitute professional help with AI tools. When it comes to safeguarding individuals who use AI for emotional well-being, however, users and non-users hold distinctly different priorities.
According to a survey conducted by Mental Health America between May 2025 and July 2026, frequent users of AI for emotional support are primarily concerned with confidentiality. A leading 22 percent of regular users believe companies must ensure no one else can see their conversations, while 12 percent want guarantees that private chats won't be used to train models. At the same time, 15 percent want companies to clearly disclose the emotional risks of using AI and 12 percent think chatbots should refer users to human professionals rather than dispensing advice. Notably, 12 percent of users would prefer the technology to feel even more human-like.
Non-users, on the other hand, take a far more cautious approach. A quarter of non-users argue that AI should strictly refer people to mental health professionals instead of giving guidance, and 19 percent emphasize the need for overt reminders that users are speaking to software, not a human. Only 11 percent of non-users advocate for industry impact standards, with an equal share wanting AI to feel less human-like.
