LLM-Based AI Chatbot Use and Short-Term Issue-Specific Stress: The Role of Perceived Message Contingency

Background/Objectives: General-purpose large language model (LLM) chatbots are increasingly used to talk through everyday stressors, yet it is unclear whether a brief conversation with such a chatbot is followed by lower stress than consulting conventional online resources, and which perceived features of the exchange are associated with stress. Methods: In a randomized online experiment, 285 U.S. adults identified a personally relevant stressor and conversed with ChatGPT (GPT-3.5) or browsed non-conversational mental health websites for about 10 min. Issue-specific stress was assessed before and after the session. Results: Stress decreased in both conditions. Contrary to our hypothesis, post-session stress adjusted for pre-session stress did not differ between conditions (adjusted difference = 0.05, 95% CI [−0.16, 0.26]); an exploratory equivalence test supported equivalence within ±0.2 SD. ChatGPT use was associated with higher perceived message contingency, which was in turn associated with greater perceived social support of all four types. Perceived message contingency was associated with lower post-session stress (indirect effect = −0.15, 95% CI [−0.30, −0.01]), but this association was offset by a positive direct path; none of the hypothesized serial pathways through the four forms of support were clearly supported. Conclusions: A brief conversation with a general-purpose LLM chatbot was followed by post-session stress comparable to that after browsing reputable mental health websites and was experienced as more contingent and, for most forms of support, more supportive. Perceived contingency may be one feature linked to lower stress, but other features of the conversation appear to have worked in the opposite direction. The findings do not speak to clinical outcomes, sustained benefits, or the replacement of professional care.