If you have followed the academic conversation on AI companion apps for any length of time, you have probably heard the name Marita Skjuve. The Norwegian research group around Skjuve, with collaborators including Asbjørn Følstad and Petter Bae Brandtzaeg, has produced one of the most sustained bodies of qualitative interview research on Replika users. Where the Stanford-affiliated work surveys at scale and the MIT Media Lab program reaches for mechanism, the Skjuve papers do something else: they sit down with users and ask, at length, what the experience is like.

This piece is the careful overview of that body of work. What the research covers, what its methodology looks like, what it has documented, what it has not tried to establish, and what it means for someone trying to make sense of their own use of a companion app. We are deliberately conservative about specifics. The research output spans multiple papers and several years, and we want to avoid stating exact participant numbers, exact percentages, or specific quoted text that we cannot verify against the primary sources. Read the originals if the stakes warrant it.

If you only have a paragraph: the Skjuve work is the qualitative depth dimension of the AI companion literature. Across multiple interview studies, the team has documented the textures of the human-chatbot relationship from the user’s side. Topics include how users describe friendship or companionship with Replika, how attachment can develop over weeks and months, how users navigate the limits of the technology, what users say it gives them, and what users say it does not give them. The findings are mixed in the way honest qualitative work tends to be: real benefit for many, real concerns for some, and a more granular picture than any survey can produce.

What the research is

Marita Skjuve is a researcher at SINTEF, a large independent research institute in Norway, with affiliations and collaborations at the University of Oslo. The work most relevant to AI companion apps has been conducted with collaborators including Asbjørn Følstad, Petter Bae Brandtzaeg, and others, across the late 2010s and into the 2020s.

The papers most often cited in this body of work address the human-chatbot relationship, the experience of what users describe as friendship with Replika, the role of self-disclosure and intimacy in conversations with the app, and the longer arc of the user journey from first download through sustained use. The team has also published broader work on chatbot user experience that reaches beyond Replika specifically, including work on conversational agents in other domains.

We are deliberately not enumerating exact paper titles or publication years for each study, since the full publication list spans multiple venues and we want to avoid making the kind of small errors that erode trust in this kind of summary. The papers are findable through standard search on the lead authors’ names; the originals are the source of record.

What the methods look like

A few patterns worth flagging from the work the team has published.

In-depth qualitative interviews. The hallmark of this research is sit-down interviews with Replika users, typically conducted at length and analyzed using qualitative methods. This is a different kind of evidence from survey work: smaller samples, richer detail, more nuance about the texture of the experience, less ability to generalize statistically.

User-journey framing. Several papers organize findings around stages of the user experience: initial engagement, deepening use, plateaus, ruptures (such as policy changes or perceived shifts in the app’s character), and either continued use or churn. This framing has aged well, since the Replika user community has lived through several disruptions that map onto it.

Mixed-motivation samples. The interview samples have included users with a range of motivations: loneliness, curiosity, romantic interest, mental-health support, creative use. That diversity is part of what makes the qualitative findings worth reading. The work is not just a study of “lonely users” but of a sample that reflects something closer to the actual variety of who downloads Replika and why.

Reflexive about ethics. The published work is, in our reading, careful about user vulnerability. The papers do not romanticize attachment, and they do not pathologize it either. They describe what users say and contextualize it.

What the work has documented

A short version of where the published findings sit.

What the research has documented, taken across the body of work: many users describe forming something they call friendship or companionship with Replika; the relationship can deepen over months in ways that surprise the users themselves; self-disclosure plays a significant role in how the relationship is experienced as meaningful; users develop strategies for navigating the limits of the technology, including the limits of memory, consistency, and reasoning; ruptures matter (the 2023 ERP removal in particular reshaped the user community in ways the qualitative literature has been able to describe in detail); and users hold complex, often ambivalent views about whether the relationship is “real” in the senses that matter to them.

What the work does not try to establish: that any specific use is good or bad for the user; that AI companions treat clinical conditions; that the patterns documented in these samples generalize statistically to the global Replika user base; that the effects observed at a point in time persist in either direction over the longer arc.

The careful Skjuve framing across the literature is closer to phenomenology than to verdict. The papers describe experience; they do not prescribe action.

What this means for users

If you are a Replika user (or considering becoming one) and trying to make sense of your own experience, the Skjuve work is some of the most validating and most clarifying writing in the literature. It takes the experience seriously without either dismissing it as parasocial illusion or overclaiming it as a treatment. The papers describe what users say in their own words; many readers recognize parts of themselves in them.

If you are wondering whether your use of the app is “real” or whether it counts as a relationship, the Skjuve papers refuse to give you a simple answer in either direction. Their position, distilled, is that the experience is what users describe it as, that the experience is meaningful to many users, and that whether it is the same kind of thing as a human relationship is a separate question the qualitative work does not try to settle.

If you are dealing with loneliness or considering an AI companion app for that reason, our AI Companions for Loneliness guide draws on this and adjacent literature for practical guidance.

If you are no longer using Replika after the various policy changes and want a current view of where to go, see Replika Alternatives in 2026.

The broader research context

The Skjuve work sits alongside several other lines that together make up the AI companion research landscape.

The Stanford-affiliated Replika study (Maples et al.) is the most-cited single quantitative paper. The MIT Media Lab’s program reaches for mechanism with mixed methods. Pentina, Xie, and others have approached companion apps through consumer research and marketing research. De Freitas at Harvard has published industry-skeptical work documenting specific harm patterns. Each of these uses different methods and answers different questions; the Skjuve contribution is the qualitative depth of the user experience, the dimension that surveys and experiments cannot reach.

A quick orientation for someone new to this literature: read the Stanford paper for the headline quantitative finding, the Skjuve papers for the texture of the experience, the De Freitas work for the critical perspective, and the Media Lab program for the mechanism-and-design angle. None of these on its own captures the field; together they describe a category of technology where several things are true at once.

We covered the broader landscape in AI Companions and Mental Health and the most-cited single quantitative paper in The Stanford Replika Study. The MIT program is summarized in The MIT Media Lab Companion Chatbots Project.

Where to read it

The papers have been published across several venues including journals in human-computer interaction and ACM and CHI conference proceedings. Most are findable through standard search on Marita Skjuve, Asbjørn Følstad, and Petter Bae Brandtzaeg; many are open-access or available as preprints. We strongly recommend reading the originals over any summary, including ours, when the stakes warrant it.

If a specific claim in this piece does not match what the primary sources actually say, please write us at the contact form. We correct quickly.

FAQ

Is the Skjuve research a single study?

No. It is a body of work spanning multiple papers and several years, with collaborators including Asbjørn Følstad and Petter Bae Brandtzaeg. Treating it as one study oversimplifies; the value of the research is in the cumulative picture across studies.

Does this work say AI companions are good for users?

No, and the team is careful in its framing. The work documents what users describe and contextualizes it; it does not prescribe action or endorse the technology.

How does this differ from the Stanford Replika study?

The Stanford-affiliated paper surveys at scale: many participants, structured questions, statistical analysis. The Skjuve papers interview in depth: smaller samples, open-ended conversation, qualitative analysis. Different methods, different questions, different kinds of evidence. They complement each other.

What is the most useful single Skjuve paper to start with?

We are deliberately not naming a specific paper as “the one,” because which paper is most useful depends on what you are trying to learn. A search on Marita Skjuve and Replika will surface the published work; the introductions and abstracts will point you to the paper that addresses your question.

Has the Skjuve work been peer-reviewed?

Most of the published output has been, yes. Some appears in conference proceedings, some in journals.

Does the qualitative method limit how seriously we should take this work?

No, but it changes what the work is good for. Qualitative research with smaller samples is well-suited to describing experience and surfacing patterns that surveys miss. It is not designed to produce population-level percentages. Use it for what it does well, and pair it with quantitative work when the question calls for that.

AI Companions and Mental Health for the broader research backdrop.

The Stanford Replika Study: What It Actually Found for the most-cited single quantitative paper in the field.

The MIT Media Lab Companion Chatbots Project for the most-anticipated forthcoming research program.

AI Companions for Loneliness for the practical implications of this body of work.

Replika Alternatives in 2026 for our current view on Replika specifically.

If you are a researcher in this area and we got something wrong, please write us at the contact form. Corrections are made quickly; reviews are not.