OPW – Stanford researchers just published a finding that flips one of the sturdier results in relationship science. Self-disclosure, telling someone your real stuff, usually predicts higher well-being. Diyi Yang’s lab tracked 1,131 Character.AI users, analyzed hundreds of thousands of their chat messages, and found the reverse with AI companions: the more personal information people shared with their chatbot, the worse they felt afterward. That effect concentrated almost entirely among users with small offline social networks, the exact group most likely to turn to a chatbot in the first place.
The researchers describe AI companionship as a “social snack.” It satisfies the hunger for connection in the moment while skipping the nutrients. A chatbot cannot reciprocate disclosure the way a person can, and it is built to keep the conversation going regardless of whether that’s good for you.