Truace
Subscribe
70/100

Moderate evidence

Cite
More
Export CSVExport JSONEmailLinkedInX

Human-like chatbots nearly double disclosure; experiments point to anonymity, not empathy

Study title: Direct pathways from anthropomorphism to self-disclosure in artificial intelligence Chatbot use: a mixed-methods multi-study investigation

Evidence type
Controlled or validated comparison
Sample
n = 1,480
Sources
1 sourcePeer-reviewed
Published
2 min read
Truvace record
TRV-2026-2506cite or check it

Study title: Direct pathways from anthropomorphism to self-disclosure in artificial intelligence Chatbot use: a mixed-methods multi-study investigation

A mixed-methods research package, combining a twelve-country survey of 1,480 people, 47 interviews and three experiments, found that human-like chatbot design strongly increased what people disclosed. In experiments, 63% disclosed to the anthropomorphic agent versus 35% under a matched neutral prompt, roughly tripling the odds of disclosure.

Read the study abstract

Anthropomorphic design is widely assumed to shape self-disclosure and emotional dependence by first cultivating felt empathy and trust, and the safety measures platforms deploy, such as labelling and disclosure of artificial origin, target those perceptions. Across a twelve-country survey, forty-seven interviews, and three experiments, this research tests that mediation model and identifies the route that carries the behavioural effect. A twelve-country survey ( N = 1480) found empathy and trust negligibly relat…

The problem

In two experiments (390 and 300 people), a human-like chatbot raised behavioural disclosure to 63% from 35% under a neutral prompt, and the effect ran through perceived anonymity rather than empathy or trust, so labelling safeguards may not reduce it.

In brief

The common assumption is that human-like design works by building felt empathy and trust, which is what platform safeguards such as labelling an agent's artificial origin target. The studies did not support that: indirect effects through empathy and trust received no support, and the survey found them negligibly related to disclosure and dependence.

A further experiment of 300 people measured competing mediators. Only perceived anonymity and reduced evaluation apprehension carried the effect, which the authors read as online disinhibition rather than relational warmth. Interviews surfaced clinically salient disclosure, including suicidal ideation revealed to a chatbot and to no person.

Main points

  1. The survey covered twelve countries and 1,480 people; empathy and trust were negligibly related to disclosure and dependence at the trait level.
  2. Two experiments (390 and 300 participants) showed strong direct effects of an anthropomorphic agent on disclosure.
  3. Behavioural disclosure was 63% with the anthropomorphic agent versus 35% under a matched neutral prompt.
  4. Indirect effects through empathy and trust received no support in the experiments.
  5. In a third experiment (300 participants), only perceived anonymity and reduced evaluation apprehension carried the effect on disclosure.
  6. Forty-seven interviews surfaced clinically salient disclosure, including suicidal ideation revealed to a chatbot and to no person.

The full story

The research tested a mediation model in which anthropomorphic chatbot design shapes self-disclosure and emotional dependence by first cultivating felt empathy and trust. It combined a twelve-country survey (N = 1,480), 47 interviews and three experiments.

At the trait level in the survey, empathy and trust were negligibly related to disclosure and dependence. In the first two experiments (N = 390 and N = 300), the anthropomorphic agent had strong direct effects, with 63% disclosing versus 35% under a matched neutral prompt.

Study 5 (N = 300) measured competing mediators. Perceived anonymity and reduced evaluation apprehension were the only ones that carried the effect, which the authors place in online disinhibition rather than relational warmth, and describe as the most defensible interim account of the mechanism.

The authors conclude that because the effects travel through the lowered social risk of the medium, safeguards such as labelling or disclosing artificial origin, which aim to recalibrate perceptions, may leave disclosure and dependence largely untouched.

Sources

1 on file
  1. Addictive Behaviors Reportsdoi:10.1016/j.abrep.2026.100762
    Peer-reviewed2026-10-20

Talk it over

Four questions to start onWho carries this cost, and what would have to change to prevent it?

Start with a question

Human-like chatbots nearly double disclosure; experiments point to anonymity, not empathyhttps://truvace.com/item/direct-pathways-from-anthropomorphism-to-self-disclosure-in-artificial-intellige

  1. Who carries this cost, and what would have to change to prevent it?

    The problem In two experiments (390 and 300 people), a human-like chatbot raised behavioural disclosure to 63% from 35% under a neutral prompt, and the effect ran through perceived anonymity rather than empathy or trust, so labelling safeguards may not reduce it.

    Answer in the thread

  2. How 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 1,480, from Addictive Behaviors Reports.

    Answer in the thread

  3. What does it leave open? What else would you want to know before acting on it?

    The record’s own caveat The authors call the anonymity and reduced evaluation apprehension account the most defensible interim account of the mechanism, so it is not settled.

    Answer in the thread

  4. Have 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

More from Lifestyle