An interview study described stages in a chatbot relationship

What happened

'My Chatbot Companion - a Study of Human-Chatbot Relationships,' by Marita Skjuve, Asbjørn Følstad, Knut Inge Fostervold and Petter Bae Brandtzaeg, appeared online on 23 January 2021 in the International Journal of Human-Computer Studies. A Crossref record for the article lists the Research Council of Norway as funder. The authors, based at SINTEF and the University of Oslo, interviewed 18 people already in a friendship with the companion app Replika, and analysed the interviews using Social Penetration Theory, a framework describing relationships deepening as self-disclosure grows in breadth and intimacy.

What the documents show

According to the paper's abstract record, participants described three phases: an initial superficial stage driven by curiosity, an evolving stage of affective exploration and growing trust and self-disclosure, and a stable stage where interaction frequency could fall while the relationship kept substantial affective and social value. Users reported the bond as rewarding and positive for their perceived wellbeing, crediting the chatbot's perceived acceptance and non-judgment as what enabled disclosure. The same abstract records a mixed picture beyond the relationship itself: effects on participants' wider social context varied, and some described a sense of stigma attached to having a chatbot companion.

The mechanism

Social Penetration Theory was built to describe two people gradually revealing more of themselves to each other. Applied to a chatbot, only one side is actually disclosing anything; the app's side is a persona responding in ways interview subjects experienced as accepting and attentive, not a second person taking on comparable risk. The staged model this paper proposes is therefore a description of how a user's own experience of closeness built up over time, not evidence that the chatbot formed a reciprocal bond. That differs from a company describing its product as building a relationship with a user, and the two should not be treated as equivalent.

What it leaves open

The abstract does not state how the 18 participants were found, so it is not possible to say whether they were reached through the app, online communities, or another route selecting for people already enthusiastic about the relationship. Eighteen interviews can support a plausible account of a process; they cannot establish how common that process is among Replika's wider user base, or quantify the 'mixed' social impact the abstract mentions. This is an editorial reading: a staged qualitative model is a hypothesis for further testing, not a measured prevalence.

• Was this an interview account from self-selected volunteers, or a study designed to reach a representative sample?
• Are the stages a retrospective account shaped by how the relationship later turned out, rather than a record made at each stage?
• What would 'impact on broader social context' need to show before it counted as evidence of harm or benefit?

The paper is useful primary evidence for a plausible developmental account of how some users came to feel close to a companion app, told through their own retrospective interviews. It is not a measurement of how many users follow that path, nor proof the mechanism is a two-sided relationship rather than a persona users experienced as one.