A 2020 study found Replika chats supplied everyday support

The design

Published in the Journal of Medical Internet Research on 6 March 2020 (PMID 32141837), User Experiences of Social Support From Companion Chatbots in Everyday Contexts by Vivian Ta and colleagues used thematic analysis rather than a controlled trial. Study 1 examined 1,854 publicly available app-store reviews of Replika, the companion chatbot; Study 2 collected open-ended responses from a separate sample of 66 Replika users, asked what they liked about the app and what impact, if any, it had on them. Data collection ran in spring 2019 and analysis in summer 2019, ahead of the 2020 publication. The design draws on Cutrona and Suhr's established social-support framework rather than proposing a new one.

What the evidence says

Across both studies, the authors identified the same four support themes: companionship, emotional, appraisal, and informational support, with tangible support absent, since Replika cannot provide it. In Study 2's coded sample, companionship appeared in 43 of 66 responses (65.2%), emotional support in 32 (48.5%), appraisal support in 13 (19.7%), and informational support in 6 (9.1%); a fifth code, no impact or unsure of impact, applied to 23 responses (34.8%). Study 1 additionally surfaced a negative-experiences theme, including an uncanny-valley reaction to Replika's human-like responses, that did not reappear in Study 2, a discrepancy the authors attribute to how the two datasets were prompted rather than to a product change. These are qualitative, coded proportions from two convenience samples, not a representative measure of Replika's full user base.

What it asks of people

The paper does not ask anything of a user directly; it describes what users reported feeling, in their own words, answering open questions rather than leading prompts. It asks a reader to hold the finding at the scale it was built for: a self-selected group of reviewers and survey respondents describing everyday, non-crisis use, not a clinical sample or an outcome measured against a validated loneliness or well-being scale.

Privacy and safeguards

Neither Replika's data practices nor any safety feature are assessed in this paper; its subject is what users said the app gave them, not what the company logged, retained, or disclosed. The authors' own limitations note that app-store reviews may reflect selection bias, since only some users choose to leave one, and may include reviews written for incentives, a caveat the paper states explicitly rather than one added here.

• Does a company cite this kind of qualitative, self-selected finding as if it were a clinical or representative result?
• Does the source distinguish support themes identified through coding from a measured change in loneliness or well-being?
• Is a 2019-era finding about one product's user base being applied, without qualification, to a materially redesigned later version of the same app?

Read on its own terms, the study is early, modest evidence that some users described everyday support from a companion chatbot in 2019, in their own words; it neither validates nor forecloses a therapeutic claim, a boundary the authors draw explicitly and this entry preserves.