When Different Relational Experiences with AI Become Visible in Conversation: Linguistic and Dyadic Evidence from Naturalistic ChatGPT Interactions
Abstract
People may experience large language models (LLMs) as therapeutically allied, emotionally close, or socially substitutive, but whether these relationships become visible in conversations is unknown. We distinguished what users discussed, what unfolded relationally, and how they experienced the AI relationship. We linked self-reports from 94 adults to donated, personally meaningful ChatGPT conversations using 21 outcome-blind frozen topics, blinded triple annotation of 500 sequences, and participant-grouped machine learning frozen before outcome access. Digital Therapeutic Alliance (DTA) and AI Attachment were related but showed different conversational correlates. Validation (R²_CV = .423) and uptake (R²_CV = .260) were recoverable out of sample, yet, neither formed a generalizable DTA signature. Conversely, a post-registration multivariate analysis identified a distributed Social Substitution fingerprint beyond conversation length and 21 topics (ΔR²_CV = .070) and improved prediction in 98% of repetitions. Ruptures were rare and consistently repaired; their association with DTA was fragile. Thus, different human–LLM relationships leave different—and sometimes no readily observable—conversational traces.
Related
- From anticipatory fear to relational trust: women's experiences of gynecological examination in Turkey: a qualitative study
- Pregnant Women's Experiences of Obstetric Abdominal Ultrasound: A Qualitative Study Exploring Emotional, Cognitive, and Relational Dimensions.
- Connectedness as a candidate pathway linking psilocybin retreat experiences to sustained wellbeing: a prospective naturalistic study
- Relational care in community mental health: Evaluating staff experiences in Intensive Community Care Services (ICCS) vs treatment as usual
- German Adaptation and Psychometric Evaluation of the Conversation Questionnaire
BibTeX
@misc{schmitt2026when,
title = {{When Different Relational Experiences with AI Become Visible in Conversation: Linguistic and Dyadic Evidence from Naturalistic ChatGPT Interactions}},
author = {Nadine Nicole Schmitt and Lennart Seizer and Fanny Alexandra Dietel and Lea Grimm-Keßler and Björn Schuller and Johanna Schuller},
howpublished = {PsyArXiv},
year = {2026},
month = oct,
doi = {10.31234/osf.io/hr7gc_v1},
url = {https://osf.io/preprints/psyarxiv/hr7gc_v1/},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - When Different Relational Experiences with AI Become Visible in Conversation: Linguistic and Dyadic Evidence from Naturalistic ChatGPT Interactions
AU - Nadine Nicole Schmitt
AU - Lennart Seizer
AU - Fanny Alexandra Dietel
AU - Lea Grimm-Keßler
AU - Björn Schuller
AU - Johanna Schuller
PY - 2026
DA - 2026/10/12
PB - PsyArXiv
DO - 10.31234/osf.io/hr7gc_v1
UR - https://osf.io/preprints/psyarxiv/hr7gc_v1/
AB - People may experience large language models (LLMs) as therapeutically allied, emotionally close, or socially substitutive, but whether these relationships become visible in conversations is unknown. We distinguished what users discussed, what unfolded relationally, and how they experienced the AI relationship. We linked self-reports from 94 adults to donated, personally meaningful ChatGPT conversations using 21 outcome-blind frozen topics, blinded triple annotation of 500 sequences, and participant-grouped machine learning frozen before outcome access. Digital Therapeutic Alliance (DTA) and AI Attachment were related but showed different conversational correlates. Validation (R²_CV = .423) and uptake (R²_CV = .260) were recoverable out of sample, yet, neither formed a generalizable DTA signature. Conversely, a post-registration multivariate analysis identified a distributed Social Substitution fingerprint beyond conversation length and 21 topics (ΔR²_CV = .070) and improved prediction in 98% of repetitions. Ruptures were rare and consistently repaired; their association with DTA was fragile. Thus, different human–LLM relationships leave different—and sometimes no readily observable—conversational traces.
ER -