The most-cited academic study of Replika is a survey of 1,006 student users published in npj Mental Health Research on January 22, 2024. It found that these users were lonelier than typical student samples yet reported high perceived social support, and 3 percent (30 people) said the app stopped suicidal thoughts. The data came from late 2021, were self-reported, and cannot show that the app caused any of these outcomes.
Key takeaways
- The study surveyed 1,006 Replika users who were students aged 18 or older and had used the app for more than a month; about 75 percent were US-based.
- Participants described using Replika as a friend, a therapist, and an intellectual mirror, often at the same time.
- Thirty participants (3 percent) reported that Replika halted suicidal ideation; this subgroup was also more likely to report depression.
- Data were collected in late 2021, before later changes to the product, and rely on self-report from people already using the app.
Who was surveyed and when
The research team, from Stanford's Graduate School of Education, invited users at random by email from a list of app users. Eligibility required being a student, being at least 18, and having used Replika for over one month. Respondents received a gift card of USD 20 for a survey that took 40 to 60 minutes. The paper states that data collection took place in late 2021 and that, at that point, Replika was not designed to initiate therapeutic or intimate relationships, although it could follow scripted therapy-style dialogs if prompted.
That timing limits what the findings say about the app today. Replika has changed since, and the paper describes the product as it was during data collection. The paper also describes the app as marketed with the line "the AI companion that cares" and says its model was co-trained with OpenAI's GPT-3 and GPT-4. This article does not verify those technical details independently.
What the survey found
The abstract reports that participants were more lonely than typical student populations while perceiving high social support. In the discussion, the authors state that 90 percent of this group reported experiencing loneliness, compared with 53 percent in prior studies of US students, and they describe the sample as typically single, young, low-income, full-time students.
Respondents used Replika in overlapping ways. The paper groups reported outcomes into companionship (availability, lack of judgment, conversational ability), therapeutic interaction, and a role as an intellectual mirror. It also found that many participants held conflicting beliefs about the app, calling it a machine, an intelligence, and a human.
| Finding | Figure reported | What it cannot show |
|---|---|---|
| Loneliness among respondents | 90 percent, against 53 percent in earlier US student studies | Whether Replika use changed loneliness over time |
| Reported suicidal ideation halted | 3 percent (30 of 1,006) | That the app caused the change, or that the effect lasts |
| Depression in that subgroup | 23 percent (7 of 30), against 6 percent (59 of 976) in the rest | Direction of cause; the subgroup is small |
Why the 3 percent figure needs care
The number appears in the abstract as a critical finding, and it is easy to quote out of context. It comes from respondents' own statements about their experience, collected once. There was no comparison group of non-users and no follow-up. People who stay with an app for more than a month, and who agree to a long survey, are also a self-selected group. Survey answers can reflect real benefit, but they can equally reflect the way a person remembers or wants to describe a period of distress.
The authors themselves call for further studies of the effects of such agents on users' mental health. Readers should treat the result as a reason to run better-designed research, not as evidence that a companion app works as a safeguard.
Reading studies like this one
- Check when the data were collected and whether the product has changed since.
- Look for a comparison group and for measures taken before and after use.
- Note who was eligible; a sample of long-term users cannot speak for people who quit.
- Separate what respondents reported from what an outside measure confirmed.
- Check who funded the work and who is affiliated with the product.
Where this leaves the question
Asking whether one app is the future of conversational AI is not something this study tests. It describes how a sample of student users in 2021 said they used and understood a companion chatbot. Regulators have looked at the same product from a different side: Italy's data protection authority fined its developer in 2025 over legal basis, transparency, and age checks, which concern data handling, not wellbeing. This article gives general information and is not medical advice; decisions about mental health care belong with a qualified clinician.
FAQ
How many people were in the Replika study?
The survey included 1,006 student users aged 18 or older, all of whom had used Replika for more than a month. About 75 percent were US-based and 25 percent international.
Did the study show that Replika prevents suicide?
No. Three percent of respondents reported that the app halted their suicidal ideation, but this is a self-reported figure from a single survey without a control group.
Is the data still current?
The data were collected in late 2021. The paper says the product at that time was not designed to start therapeutic or intimate relationships, so later versions may differ.
Where was the study published?
It appeared in npj Mental Health Research on January 22, 2024, under the title "Loneliness and suicide mitigation for students using GPT3-enabled chatbots."
Related reading
- Replika and the Italian privacy fine
- AI friend chat apps
- How conversational interfaces developed
- The Grok chatbot prototype
More in AI Tools and Industry.
Last verified: October 2026
Sources
- Loneliness and suicide mitigation for students using GPT3-enabled chatbots, Maples, Cerit, Vishwanath, and Pea, npj Mental Health Research, 2024
- AI: the Italian Supervisory Authority fines company behind chatbot Replika, European Data Protection Board, May 21, 2025
- Europe PMC record for the paper, European Bioinformatics Institute, accessed October 2026