Between Help and Manipulation: Youth Recognition of Dark Patterns and Value Tensions in LLM-Assisted Mental Health Support - A Qualitative Think-Aloud Study
Exploratory qualitative study in which 19 participants aged 18 to 24 compared ten pairs of scenario-based chatbot responses about mental health in think-aloud interviews. Each pair set a neutral response against one embedding dark patterns such as simulated authority, anthropomorphic self-presentation, simulated intimacy, sycophantic agreement, discouragement of help-seeking, fabricated social consensus, emotional retention, interaction padding, therapeutic gamification and disguised advertising. Reflexive thematic analysis mapped which patterns young people recognise, how, and the value tensions they navigate.
Publisher
JMIR Preprints (JMIR Publications)
Published
16 Aug 2026
Added
today
Key Findings
- Overt commercial and coercive cues (guarantees, subscription-style promotion, pressure to continue, gamification, claims of superior intimacy, discouragement of professional help) were recognised most readily
- Recognition followed three patterns: explicit recognition, intuitive discomfort without explanation, and reflective recognition after consideration
- Anthropomorphic language, social proof, reassurance and clarifying questions were ambiguous because some participants experienced them as warm or useful
- Several participants anticipated accepting responses they considered inaccurate or manipulative when distressed and seeking reassurance
- Four value tensions: personalisation versus privacy, safety versus perceived utility, accessibility versus quality of care, warmth versus over-humanisation; epistemic versus emotional trust and transparency versus reassurance cut across them
Methodology Notes
n=19, ages 18 to 24; one-to-one online semi-structured think-aloud interviews; scenario and response order randomised; reflexive thematic analysis with deductive sensitising concepts from a dark-pattern and human-values framework. Preprint under JMIR review; author affiliations are not in the Crossref record and the preprint page returns a challenge shell to automated fetches, so they were not confirmed. Posted 16 August 2026 (Crossref posted date); the Crossref record was created 17 September 2026, which is why it surfaced this window. Verified through the Crossref record's structured abstract (4,511 characters).
Topics
Authors
Roth, Kerstin, Zhong, Vivienne Jia, Jacob, Christine
Tags
Cite This
APA
Roth, Kerstin, Zhong, Vivienne Jia, Jacob, Christine. (2026). Between Help and Manipulation: Youth Recognition of Dark Patterns and Value Tensions in LLM-Assisted Mental Health Support - A Qualitative Think-Aloud Study. JMIR Preprints (JMIR Publications). https://preprints.jmir.org/preprint/109536
Related Insights
Dark Patterns in AI Chatbots: A Taxonomy to Inform Better Design
Center for Democracy & Technology (CDT Research) · 29 May 2026
When AI Says "I Am Unable to Answer": Understanding User Responses to AI Refusals
arXiv (The Pennsylvania State University; Seoul National University, Center for Trustworthy Artificial Intelligence) · 14 Sept 2026