JMIR Publications

ABSTRACT

Background:

Psychological distress and unmet psychosocial needs are common across the cancer continuum. Conversational artificial intelligence (AI) agents may offer scalable support, but their technologies, support functions, evaluation approaches, and safety governance remain fragmented.

Objective:

To map evidence on conversational AI agents used for psychological support among patients with cancer and cancer survivors, focusing on technologies, support content, evaluation outcomes, and safety, ethics, and governance.

Methods:

A scoping review was conducted in accordance with established scoping review methodology and reported following PRISMA-ScR. Thirteen English and Chinese databases and additional registers were searched from inception to 5 May 2026. Eligible sources involved patients with cancer or cancer survivors, examined AI-based conversational agents in real-world or simulated patient-facing oncology contexts, and reported psychological, emotional, psychosocial, coping-related, survivorship-related, supportive-care, or quality-of-life relevance. Data were charted using a four-domain framework covering technologies, support content, evaluation outcomes, and safety governance.

Results:

Twenty-four studies published between 2020 and 2026 were included. Large language model or generative AI systems were the most common system form (n=8, 33.3%), followed by rule-based, knowledge-based, retrieval-based, or decision-tree conversational agents (n=7, 29.2%). Most studies included or examined information, education, navigation, or resource recommendation (n=22, 91.7%), emotional or disclosure support (n=19, 79.2%), and symptom self-management or quality-of-life support (n=13, 54.2%). Structured psychological or behavioural intervention components were less common (n=8, 33.3%). Evaluation most frequently addressed feasibility, usability, acceptability, satisfaction, or user experience (n=15, 62.5%), whereas standardized psychosocial outcomes and long-term follow-up were less consistently assessed. Ethics-related procedures were commonly reported (n=21, 87.5%), but crisis escalation or referral (n=8, 33.3%), hallucination, bias, or misinformation management (n=9, 37.5%), AI identity transparency (n=11, 45.8%), and role-boundary or non-replacement statements (n=6, 25.0%) were less consistently described.

Conclusions:

Conversational AI for cancer psychological support is an emerging, heterogeneous, and not yet clinically mature field. Current evidence is concentrated in early-stage, usability-oriented, and technology-diverse studies. Future research should specify theory-informed psychosocial mechanisms, evaluate standardized outcomes over time, integrate nurse or clinician oversight, and report safety-governance pathways before wider implementation in psycho-oncology care. Clinical Trial: none