Conversational AI is increasingly used to rehearse and articulate distress before people seek formal
mental health support. This review examines evidence across three clinically relevant domains: selfdisclosure
to conversational systems, stigma-related barriers that shape uptake, and the conditions
under which AI use supports movement from private coping toward real-world care pathways. Findings
suggest disclosure in chatbot contexts is sensitive to perceived anonymity, privacy expectations, trust
cues, and fear of judgement, and may provide short-term emotional relief. Self-stigma and label
avoidance are associated with attitudes toward AI-delivered support, potentially lowering the threshold
for initial engagement while also increasing the risk of avoidance maintenance. Early feasibility research
indicates that screening and referral chatbots can be acceptable and may reduce navigation burden,
and outcome evidence shows that some structured conversational interventions can reduce distress in
specific populations. The review consolidates these results into a mechanism-led synthesis of where AI
use is most likely to assist early help-seeking, where it can stall, and the bridge conditions that increase
the probability of transfer to human or formal supports.
