Do AI Companions Understand? Most UK Teens Say Yes
An online survey of 1,009 UK teenagers aged 13–18 who had used AI companions: which platforms they use, why, how much they trust the advice, and what they believe these systems can do.
What it contributes
The central finding is attributional: 56% of teen users believe AI companions can think, while 77% say they cannot feel. ChatGPT dominates companion use (78%), far ahead of purpose-built companion apps — teens use mainstream assistants relationally. 31% interact most days; 52% report having chosen, at least once, to discuss something serious or important with an AI companion instead of a person; 53% express moderate-to-complete trust in the advice. Respondents' perceptions of the social impact cut both ways: 67% report no effect on their human friendships and 26% say AI companions help them make human friends, while 32% find AI conversations more satisfying than those with friends and 7% say AI is replacing some friendships. The authors offer emulated empathy as their interpretation of the mechanism of influence, argue for governance that addresses relational dynamics rather than product categories, and explicitly caution against moral panic.
Read with care
A single online self-report survey of existing users — non-users were screened out, so it cannot estimate prevalence among all teenagers, and 'representative' is the authors' description with limited sampling and weighting detail published. The survey measures reported experiences and beliefs; it does not establish that companions cause harm or benefit, nor that their advice is accurate. The sample skews slightly female and younger, as the authors note. The report states two slightly different fieldwork windows (7–17 November and 27 October–7 November 2025). A lab report funded by Responsible AI UK, not a peer-reviewed publication; the free-text clustering method is not detailed.
Why it belongs here
A useful UK snapshot of teenage AI companion users — and strong support for the point that companion dynamics live inside general-purpose assistants, not only dedicated apps. Directly relevant to Online Safety Act debates and to anyone designing for or assessing young users.
Report of the Emotional AI Lab, Bangor University, published 2 January 2026; funded by Responsible AI UK. Peer-reviewed publication not claimed.
Source reading depthFull 20-page report read, including policy recommendations; percentages checked against the report's text, tables and charts.
Source check: 20 October 2026. Reading cautions are editorial interpretation, not quotations from the authors.