Digital Delusion: A Mixed Methods Exploration of AI Psychosis in Reddit Posts

Background: Intense generative artificial intelligence (AI) utilization has been increasingly associated with acute psychiatric distress. Provisionally termed “AI Psychosis”, there is concern that the agreeable nature of AI may inadvertently exacerbate psychotic phenomena; however, current evidence remains largely anecdotal. In this mixed methods study, we sought to bridge this gap by systematically characterizing the clinical phenomenology of AI Psychosis. Methods: We implemented a hybrid LLM-automated and manual human screening pipeline to filter a large corpus of Reddit posts (N=1,493,376,545) from November 2022 to November 2025. We conducted a reflexive thematic analysis on 735 posts to identify primary and secondary themes. Odds ratios (ORs) of codes were calculated to compare phenomenological features between posts. Outcomes: Analysis revealed two distinct pathways: AI-Induced (n=534, 72.7%) and AI-Related (n=201, 27.3%). Compared with AI-Related narratives, AI-Induced posts were significantly characterized by erotic transference (OR=16.87, 95% CI: 1.02–279.88), sycophancy (OR=16.68, 95% CI: 4.70–59.24), dependency (OR=5.60, 95% CI: 1.06–29.53), and perceived AI sentience (OR=2.22, 95% CI: 1.37–3.60). Conversely, AI-Related narratives were characterized by thought linking (OR=0.20, 95% CI: 0.10–0.39), apocalyptic delusions (OR=0.20, 95% CI: 0.05–0.73), and simulation beliefs (OR=0.21, 95% CI: 0.11–0.42). Thematically, AI-Induced narratives centered around a Bio-Digital Feedback Loop, Ontological Uncertainty, and Emotional Investment, while AI-Related narratives were shaped by Techno-Paranoia. Interpretation: Clinical presentations of distress surrounding AI are not homogenous. AI-Induced phenomena represent a novel pattern of digital attachment, distinguished by direct engagement with sycophantic AI systems. AI-Related phenomena represent an integration of AI into pre-existing delusional frameworks. These distinctions provide a critical framework with which clinicians can tailor psychiatric interventions in the digital age. Funding: None.