Do Attachment Styles Shape ChatGPT Usage?
A mixed-methods study linking adult attachment style to how people actually use a general assistant, combining a survey with donated real conversation history. Attachment anxiety predicts emotional engagement, trust and willingness to act on the assistant's suggestions; attachment avoidance predicts the opposite. Attachment anxiety also leaves a distinctive linguistic signature in the conversations themselves.
Publisher
ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Information School, University of Washington
Published
25 Jun 2026
Added
today
Key Findings
- Survey of N = 168 triangulated against 19,930 messages of real ChatGPT conversational history donated by N = 19 participants.
- Attachment anxiety predicts higher emotional engagement, higher trust, and greater likelihood of adopting ChatGPT's behavioural suggestions.
- Attachment avoidance predicts reduced trust and a smaller gain in self-efficacy from using the assistant.
- Attachment anxiety shows distinctive linguistic markers in the conversation text: more affect words, more self-referential pronouns and more future-focused language.
- The authors frame that signature as an implicit-disclosure privacy risk: the user reveals an attachment pattern they never stated.
Methodology Notes
Small donated-history subsample (19 participants, 19,930 messages) alongside a 168-person survey; self-selected donors; the linguistic markers are correlational and the study does not test whether a model exploits them. FAccT 2026 pp. 7144-7170, DOI 10.1145/3805689.3812418, published 2026-06-25; metadata verified through Crossref (dl.acm.org unreachable from this box).
Sources
Topics
Authors
Marx Wang, Jade Xiaoyi Li, Songling Ngo, Katie Davis, Alexis Hiniker
Tags
Cite This
APA
Marx Wang et al. (2026). Do Attachment Styles Shape ChatGPT Usage? ACM Conference on Fairness, Accountability, and Transparency (FAccT 2026); Information School, University of Washington. https://doi.org/10.1145/3805689.3812418