AI-Mediated Mental Health Support on Reddit: Annotated Dataset of 5,126 Posts Across 47 Communities

This dataset contains 5,126 Reddit posts describing experiential or exploratory use of AI systems for emotional support and therapy-related purposes. Posts were collected from 47 mental health subreddits between November 30, 2022 and August 15, 2025. The data were identified from a large corpus using a multi-stage filtering pipeline combining keyword retrieval, LLM-based validation, and human review. Only posts meaningfully discussing AI use in mental health contexts were retained. Each post is annotated using a theory-grounded framework based on the Technology Acceptance Model and therapeutic alliance theory. Labels capture adoption attitudes, perceived usefulness, trust, output quality, risks, task and goal alignment, bond, sentiment, usage intent, and comparison to human therapy. To reduce re-identification risk, raw post text is not included. The dataset provides post identifiers and derived annotations in accordance with platform policies and ethical research practices. If you use this dataset, please cite the associated paper: Aghakhani, E., & Rezapour, R. (2026). Like a Therapist, But Not: Reddit Narratives of AI in Mental Health Contexts. arXiv preprint arXiv:2601.20747.