Peer-reviewed & preprints
Research
Academic and clinical scholarship on how conversational AI affects the people who use it — from crisis-response performance to companionship, dependency, and psychosis.
356 entries, newest first
Preprint
Mitigating Social Sycophancy via Pluralistic Preference Optimization
Proposes Pluralistic Preference Optimization (PlurPO), a post-training method in which a model simulates the stakeholders affected by a user's interpersonal situation and is trained to prefer respons…
Peer-reviewed
Artificial Intelligence–Associated Psychosis
A psychiatric case report from the mental health division of Northern Hospital, Melbourne, describes a 27-year-old woman with a year of psychotic symptoms whose presentation centred on ChatGPT. She u…
Preprint
Detecting Suicide Risk with AI Chatbots: Real-World Performance Within a Clinically Supervised Workflow
A retrospective cohort study by Spring Health evaluates an LLM-based safety agent (gpt-4o, prompted with C-SSRS and SAFE-T frameworks) that classifies suicide risk into four levels during a five-minu…
Peer-reviewed
Programmed Interventions To Prevent Delusions From Excessive Use of Conversational AI Bots
A letter in ACM AI Letters from the Georgia Institute of Technology tests whether prompting interventions designed to reduce sycophancy also reduce large language models' endorsement of users' delusi…
Preprint
Persona and Persuasive Framing in AI Voice Agents: A 2x2 Field Experiment with Children
Randomised 2x2 field experiment embedded in a public German-language Santa Claus telephone hotline before Christmas 2025, in which children's calls were routed to LLM voice agents that varied persona…
Peer-reviewed
AI or Human Support for Suicide Prevention? Examining Help-Seeking Intention in Suicidal Crisis
A cross-sectional web survey of 1,024 Japanese adults aged 18-69 with severe psychological distress (Kessler-6 score of 13 or more) asks whether they would use chat-based crisis support delivered by…
Preprint
How People Use ChatGPT: Conversation-Level Evidence from India, Nigeria, Brazil, and Pakistan
Data-donation study of complete ChatGPT exports from 1,252 users in India, Nigeria, Brazil and Pakistan (202,590 conversations, December 2022 to February 2026), paired with self-reported age and gend…
Peer-reviewed
AI-Mediated Mental Health Support: The Role of Attachment Orientation and Psychological Distress
A preregistered cross-sectional survey of 584 Israeli adults who use general-purpose generative AI asks whether attachment orientation (ECR-RS) and psychological distress (DASS-21) are associated wit…
Preprint
From Symptom Networks to Conversation Networks: A Cross-Sectional Study Mapping the Topology of Suicide-Related Clinical Dialogue
The study applies network analysis to suicide-related content in 110 German-language psychotherapy interview transcripts from the SPEAK-SAFE study. The open-weights Qwen3-32B model classified utteran…
Preprint
Right Words, Wrong Moment: A Clinician-Grounded Analysis of Distress in 19,930 Conversations between Young People and ChatGPT
Mixed-methods study of how young adults aged 18 to 25 use ChatGPT when distressed, built on complete donated ChatGPT histories (19,930 conversations from 158 participants) plus a survey that included…
Peer-reviewed
Initial Psychometric Evaluation of the Problematic Use of Generative Artificial Intelligence Scale Among Chinese School Students
A school-based cross-sectional survey evaluates a scale for problematic use of generative AI (PUGAIS, adapted from the Smartphone Addiction Scale-Short Version) among 19,484 students in grades 4-9 in…
Peer-reviewed
Making medical AI benchmarks clinically interpretable: the case of mental health
Argues that general medical AI benchmarks should report domain-specific results, and demonstrates this on HealthBench by isolating its mental health conversations. The authors compare mental health s…
Preprint
Medical Knowledge Is Not All You Need: When Medical Q&A Becomes Situated Patient Assistance
Study of the questions skin cancer patients asked a voice assistant while practising postoperative wound care, followed by an offline replay of those questions to general-purpose language models. Man…
Peer-reviewed
Beyond Replacement: Episode-Based Pathways Between Generative AI and School Counselor Help-Seeking Among Indonesian Adolescents
Explanatory sequential mixed-methods study of 432 Indonesian students aged 15 to 18. It maps how GenAI and human support, including school counselors, are used within the same personal or emotional p…
Peer-reviewed
One Question, Four Voices: How Advice for Alzheimer's Caregiving Differs Between Caregivers, Clinicians, and Large Language Models
A side-by-side comparison of responses to 85 real caregiver questions from the ALZConnected forum across peer caregivers, physicians, ChatGPT and CareGPT, a retrieval-augmented variant grounded in pe…