From Chatbots to Confidants: A Cross-Cultural Study of LLM Adoption for Emotional Support
Natalia Amat-Lefort, Mert Yazan, Amanda Cercas Curry, Flor Miriam Plaza-del-Arco · 2026
Read From Chatbots to Confidants: A Cross-Cultural Study of LLM Adoption for Emotional Support by Natalia Amat-Lefort, Mert Yazan, Amanda Cercas Curry, Flor Miriam Plaza-del-Arco (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-25525.txt.
Introduction
Amat-Lefort and three colleagues present the first cross-cultural study of LLM emotional support use. LLMs are used as always-available confidants. They survey 4,641 people in seven countries. Adoption varies sharply, from 20 percent to 59 percent. Mixed models separating cultural from demographic effects find age 25 to 44, religious, married, and higher socioeconomic status predict positive views, with status strongest. English-speaking countries show more positive views than Continental European countries. They also collect 731 multilingual prompts about loneliness, stress, relationship conflicts, and mental health.
Catalog record
| CREATOR | Natalia Amat-Lefort, Mert Yazan, Amanda Cercas Curry, Flor Miriam Plaza-del-Arco |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 66893 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.25525
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
From Chatbots to Confidants: A Cross-Cultural Study of LLM Adoption for Emotional Support Natalia Amat-Lefort, Mert Yazan, Amanda Cercas Curry, Flor Miriam Plaza-del-Arco Abstract Large Language Models (LLMs) are increasingly used not only for instrumental tasks, but as always-available and non-judgmental confidants for emotional support. Yet what drives adoption and how users perceive emotional support interactions across countries remains unknown. To address this gap, we present the first large-scale cross-cultural study of LLM use for emotional support, surveying 4,641 participants across seven countries (USA, UK, Germany, France, Spain, Italy, and The Netherlands). Our results show that adoption rates vary dramatically across countries (from 20% to 59%). Using mixed models that separate cultural effects from demographic composition, we find that: Being aged 25-44, religious, married, and of higher socioeconomic status are predictors of positive perceptions (trust, usage, perceived benefits), with socioeconomic status being the strongest. English-speaking countries consistently show more positive perceptions than Continental European countries. We also collect a corpus of 731 real multilingual prompts from user interactions, showing that users mainly seek help for loneliness, stress, relationship conflicts, and mental health struggles. Our findings reveal that LLM emotional support use is shaped by a complex sociotechnical landscape and call for a broader research agenda examining how these systems can be developed, deployed, and governed to ensure safe and informed access. https://confidant-ai.github.io/LLMs-emotional-support/ 1 Introduction Large Language Models (LLMs) are a jack of all trades, performing diverse tasks such as question answering and decision support Chatterji et al. 2025 . Beyond these instrumental uses, researchers and designers are increasingly interested in AI systems as potential companions or sources of interpersonal support ( Andersson 2025 ; Wu et al. 2025 ; Savoldi et al. 2025 , e.g.) . In particular, conversational interfaces make LLMs accessible for emotionally oriented interactions such as seeking reassurance, discussing personal concerns, or reflecting on experiences 1 1 1 https://hbr.org/2025/04/how-people-are-really-using-gen-ai-in-2025 . Figure 1: Global and Demographic Landscape in LLM Adoption and Perceptions for Emotional Support (N=4,641). SES: Socio-economic status. At the same time, many individuals face barriers to accessing traditional mental health support, including cost, stigma, and limited availability of services Hou et al. 2025 ; Tyson et al. 2023 . As a result, digital technologies are increasingly explored as complementary sources of emotional support Chaudhry and Debi 2024 ; Pandya et al. 2024 . Conversational AI systems may lower barriers to seeking help by offering immediate, and always-available interaction Haensch 2025 ; Pandya et al. 2024 . Yet, the growing use of LLMs for emotionally sensitive interactions raises questions about safety, trust, and appropriate system design Kaffee et al. 2025 . While these ethical and design implications have received significant attention, empirical evidence remains scarce on how commonly the general public turns to LLMs for emotional support and what factors shape these interactions. In particular, we lack population-level evidence on (1) who adopts LLMs for emotional support, (2) how sociodemographic and cultural backgrounds shape their perceptions of these systems, and (3) what users actually disclose to these systems. To address this gap, we conduct a survey examining how members of the public report using LLMs for emotional and mental health–related support. We examine whether cultural and sociodemographic factors mediate the use of LLMs for emotional support (Figure 1 ). Specifically, we address three pivotal research questions (RQs): (RQ1) Who uses LLMs for emotional support, and what are the main purposes and benefits that motivate people to use them? (RQ2) How do demographics and cultural background shape user perceptions of these systems? (RQ3) What do users actually disclose when seeking such support? Contributions 1) We present the first cross-cultural empirical study of GenAI chatbot use for emotional support, surveying 4,641 participants across seven countries. 2) We provide an extensive sociodemographic and cultural analysis using Cumulative Link Mixed Models across three constructs (Trust and perceived privacy, Perceived benefits, and Usage intention) to disentangle demographic from cultural effects. 3) We collect and analyze 731 real user prompts across six languages, identifying key emotional support topics. Together, our findings provide the first population-level evidence of how cultural and sociodemographic factors shape human–AI emotional support interactions, highlighting a complex sociotechnical landscape that motivates a broader research agenda to explore how these systems can be developed, deployed, and governed to promote use that is safe and informed. 2 Related Work Interest in public perceptions of conversational AI for wellbeing has grown considerably, prompting surveys, usability studies, and social media analyses, predominantly conducted in English Haensch 2025 ; Chaudhry and Debi 2024 ; Cameron et al. 2018 . Findings suggest that users value the availability, non-judgmental nature, and accessibility of AI conversational agents, while pointing to shallow and context-insensitive responses as key limitations Chaudhry and Debi 2024 ; Haensch 2025 . Younger adults and those with higher education and AI familiarity tend to show relatively more openness toward AI tools Tyson et al. 2023 . These attitudinal patterns are consistent with the CASA paradigm, which predicts that people apply the same social norms to AI as to humans Nass and Moon 2000 and disclose more under conditions of reduced social presence Papneja and Yadav 2025 . The psychological barriers underlying AI resistance, including perceived opacity, emotionlessness, and rigidity, further shape whether individuals are willing to engage with these systems De Freitas et al. 2023 . Beyond wellbeing, broader surveys of general LLM use show that SES and demographic factors stratify adoption rates and interaction styles, with higher-SES users interacting more abstractly Bassignana et al. 2025 and younger, more educated users showing higher adoption and literacy Savoldi et al. 2025 . Yet none of these works study the adoption and user perceptions of LLMs specifically for emotional wellbeing support, and all remain confined to a single country and predominantly Anglophone populations, leaving open whether adoption and perception patterns vary across cultural and sociodemographic contexts. We address this gap by surveying 4,641 participants across seven countries in six languages, collecting prompts from real prior interactions, and providing the first cross-cultural analysis of LLM adoption for emotional support that explicitly separates demographic from cultural effects. 3 Survey Methodology 3.1 Study Design We conduct a cross-cultural survey to examine demographic and cultural variation in the adoption and perception of LLMs for emotional support. We select seven countries representing the highest global shares of ChatGPT 2 2 2 Currently the most used conversational LLM. visitors FirstPageSage 2026 : the United States (17.1%), France (4.3%), Spain (3.7%), the United Kingdom (2.7%), Italy (2.5%), Germany (2.4%), and the Netherlands (1.1%), ensuring our sample reflects the populations most actively engaging with these systems. Between November 11 and 22, 2025, we collected the data via a Qualtrics panel, 3 3 3 Qualtrics provides reliable access to diverse, cross-cultural respondents and robust built-in data quality checks. using quota sampling to ensure even distribution across countries (N=200 per country) and gender balance within each national cohort. Before starting the survey, we presented participants with an onboarding screen to obtain informed consent. To establish a shared baseline understanding, we provided examples of “AI agents” (e.g., ChatGPT, Gemini, Claude) and defined “emotional support and mental wellbeing” as using these tools for managing stress, processing emotions, or personal reflection. We informed participants of the survey’s anonymous, voluntary nature,