aceModerate evidence
More
Chatbots that share personal details back got users to open up more, not to trust them more
Study title: Activating Social Responses in Human–AI Interaction: The Effects of Reciprocal Self-Disclosure and Social Role Framing
- Evidence type
- Controlled or validated comparison
- Sample
- n = 169
- Sources
- 1 sourcePeer-reviewed
- Published
- 1 min read
- Truvace record
- TRV-2026-2659cite or check it
Study title: Activating Social Responses in Human–AI Interaction: The Effects of Reciprocal Self-Disclosure and Social Role Framing
An experiment with 169 participants tested how LLM-based chatbots draw social responses from people. Participants were split across a 2 × 3 design: the chatbot either reciprocated self-disclosure or did not, and it was framed as a mentor, a companion or an assistant.
Read the study abstract
As chatbots become increasingly embedded in everyday communication, understanding when they elicit social responses is important. We conducted an experiment in which 169 participants interacted with LLM-based chatbots in a 2 (reciprocal self-disclosure: yes vs. no) × 3 (social role: mentor, companion, or assistant) between-subjects design. Self-esteem was measured before the interaction as an individual-difference factor. Factorial analyses showed that reciprocal self-disclosure increased users’ disclosure but d…
The problem
In an experiment with 169 participants, a chatbot that reciprocated self-disclosure did not significantly raise users' trust or perceived support, and users reported lower enjoyment than with a non-reciprocating chatbot.
In brief
Reciprocal self-disclosure led users to disclose more. It did not significantly increase trust or perceived support, and enjoyment was lower when the chatbot reciprocated.
Role framing worked differently depending on the person. Social role interacted with self-esteem in predicting the quality of self-disclosure, with favorable responses to the mentor role among participants with higher self-esteem.
Main points
- The experiment involved 169 participants who interacted with LLM-based chatbots.
- It used a between-subjects design crossing reciprocal self-disclosure (yes or no) with three social roles: mentor, companion or assistant.
- Reciprocal self-disclosure increased how much users disclosed.
- Reciprocal self-disclosure did not significantly increase trust or perceived support.
- Enjoyment was lower in the reciprocal-disclosure condition.
- Participants with higher self-esteem responded favorably to the mentor role when it came to the quality of their self-disclosure.
The full story
Participants were randomly placed in one of six conditions combining reciprocal self-disclosure and a chatbot social role. Self-esteem was measured before the interaction as an individual-difference factor.
Factorial analyses showed that a chatbot sharing about itself changed what users did, since they disclosed more, but not how they judged the chatbot on trust or support. Enjoyment was lower in the reciprocal condition.
The authors conclude that chatbot social cues elicit different behavioral and evaluative responses, and that role framing should account for user characteristics. They describe the study as offering a conditional account of social responding in human–AI interaction, with implications for chatbot design in relational contexts.
Sources
1 on file- International Journal of Human–Computer Interactiondoi:10.1080/10447318.2026.2738478
Talk it over
Who carries this cost, and what would have to change to prevent it?
Chatbots that share personal details back got users to open up more, not to trust them morehttps://truvace.com/item/activating-social-responses-in-human-ai-interaction-the-effects-of-reciprocal-se
Who carries this cost, and what would have to change to prevent it?
The problem In an experiment with 169 participants, a chatbot that reciprocated self-disclosure did not significantly raise users' trust or perceived support, and users reported lower enjoyment than with a non-reciprocating chatbot.
Answer in the threadHow do we know? What can this kind of evidence show, and what can’t it?
The evidence This rests on a controlled comparison with a sample of 169, from International Journal of Human–Computer Interaction.
Answer in the threadWhat does it leave open? What else would you want to know before acting on it?
Answer in the threadHave you seen AI used this way where you study, work or live? Did it go the same way?
Answer in the thread
The questions are the same on every record. The details under them are taken from this record as written.
What readers say