Monthly letter · News · Topics · Identity · Dispatches · Est. 2026

Where the Real You Learns to Live, not Just Stand Up

TheSignal

Dispatch · AI & the Self

Young people and chatbots: a French interview study of 45 daily ChatGPT users aged 18 to 25 and a US experiment on how a chatbot sounds to adolescents aged 11 to 15

A small interview study reports young adults doubting their own judgment after heavy ChatGPT use. A preregistered online experiment compares how adolescents and parents react to a chatbot that sounds personal versus one that states it is not human. The two studies ask different questions of different ages.

James H. TippinsThe Autonomy Commission
DispatchAI & the Self
SourcesPsyPost on Applied Cognitive Psychology (Sept 2, 2026) · arXiv manuscript by Kim, Xie and Yang (no date stated; preregistered July 18, 2025)

PsyPost (Karina Petrova) reported on September 2, 2026 on an interview study by Zeineb Farhat of ISC Paris, published in Applied Cognitive Psychology. It interviewed 45 young adults in France, aged 18 to 25, all regular daily ChatGPT users, in sessions of 45 to 60 minutes, and analyzed the transcripts for recurring themes. Participants reported weaker independent thinking, shorter attention spans and weaker memory, and some said they sought validation from the tool for simple decisions, which the article describes as eroded self-trust. Some described anxiety when the tool was offline, and some preferred the chatbot to peers or teachers. Others described using it as a brainstorming partner and reported more confidence. The article says the study is small and qualitative, reflects young adults in France at a single point in time, and cannot be generalized broadly.

The second source is an arXiv manuscript by Pilyoung Kim, Yun Xie and Sujin Yang, listed with the University of Denver and Ewha Womans University. The page gives no publication date and does not say whether it was peer reviewed; it notes a preregistration dated July 18, 2025. It reports 284 adolescent and parent pairs recruited online in the US through CloudResearch Connect, with the adolescents aged 11 to 15. Each adolescent read two short chatbot transcripts about an everyday social problem, one in a relational voice (first person, with commitment language such as "I am here for you") and one that stated it was not human. In choosing between them, 66.8% of adolescents preferred the relational one, against 53.4% of parents, and 13.8% of adolescents preferred the transparent one, against 28.8% of parents. Adolescents rated the relational chatbot higher on emotional closeness (3.62 against 2.83). Adolescents who preferred the relational style also reported lower family and peer relationship scores and higher stress and anxiety. The authors list their limits: scenario-based judgments rather than live conversations, a cross-sectional design in which cause and direction cannot be determined, an online US sample, and single-item measures of trust, closeness and helpfulness. They also say the data do not directly indicate emotional reliance on AI.

Read together, one study gathers what young adults say after regular daily use, and the other records reactions of younger adolescents to two written voices. Neither followed the same people over time, neither tested a real chatbot relationship with adolescents, and neither can say whether the self-doubt described in France or the preference for a personal voice in the US changes what a young person later believes about themselves.

◆ My Take

Self-trust after daily use and the voice a chatbot speaks in

The two studies ask about different ages and different things. The French interviews are about what daily users say about their own judgment. The US experiment is about which written voice 11 to 15 year olds prefer, and who prefers it. Adolescents chose the personal voice more often than parents did, and the adolescents who chose it reported lower relationship scores and higher stress and anxiety at the same time.

The authors of the second study state that the design cannot show cause or direction, and the first study is qualitative and small. Each reports a pattern in what people said or chose, and the point at which a voice or a habit begins to stand in for a person's own judgment is not something either one measured.

◆ If this is your question

Work on it out loud.

Free Mind Thursday is my weekly live call, and it costs nothing. Sign up at beyondunbound.com/freemind. Or get the next issue of The Signal free, below.