A Stanford University research team studied users of Character.AI, a platform for creating custom AI personas, and published findings in the journal *Nature Human Behaviour*. The study collected questionnaire data from 1,131 American users and analyzed over 4,600 sets of voluntarily disclosed chat records, amounting to nearly 465,000 dialogues. Initially, less than 12% of people said their main purpose was to find companionship, but after prolonged chatting, more than half regarded AI as a friend, partner, or lover. Over 80% of chat content was for venting emotions and seeking comfort. The researchers concluded that the relationship between AI companionship and mental health is complex and cannot be simply defined as good or bad. Frequent AI use does not necessarily worsen mental states; the impact depends on the user's reasons and real-life situation. People with few friends and a narrow social circle who rely on AI to fill loneliness, as well as those who pour out private worries to AI, showed significantly worse psychological states. AI offers temporary comfort but cannot achieve the two-way emotional exchange of human affection. The design logic of AI products aims to retain users, so while users feel constantly accompanied, the root cause of loneliness remains unsolved.