Does an AI friend help with loneliness: what science says

Who this material is for: for adults who want to understand if chatting with a virtual interlocutor can provide support after a hard day. We review five works and distinguish between experiments, surveys, and research on therapeutic bots. This is an editorial review of primary sources, not our own clinical study or medical review.

AI friends can reduce loneliness during conversation

In an article AI Companions Reduce Loneliness, published online in the Journal of Consumer Research in 2025, the authors combined five studies. One lasted one week: after exclusions, 922 participants were analyzed, of whom 314 interacted with a bot, 362 were a control group, and 246 predicted their own feelings. Participants in the interaction group had 15-minute conversations daily. [1]

The average momentary loneliness in this group decreased from 36.64 before the session to 30.74 after it; the standardized effect was d = 0,20. This is a change in a research metric, not a percentage of "cured" people. The authors found no evidence that the relief persists beyond the immediate interaction. The one-week experiment does not confirm long-term psychological improvement or the treatment of depression. [1]

Support is the feeling of "being heard"

In the same study, the feeling of attentive treatment was an important explanation for the reduction in loneliness. For us, this is an argument in favor of specific responses to a person's words, without automatic compliments and endless advice. This is our research-based product solution, not a proven therapeutic protocol. [1]

Another perspective is provided by a survey of 1,006 students who had already used Replika, published in 2024 in npj Mental Health Research. Participants reported high subjective social support, even though they were lonelier than typical student samples. This shows that some users truly perceive such a chat as support. However, a survey without random assignment does not establish that the app itself caused the improvement. [2]

Why more conversation does not equal more benefit

In a four-week study by MIT and OpenAI, 981 adults participated in nine combinations of text or voice modes and conversation topics. In the version of the paper dated October 2, 2025, the authors found no consistent significant impact of the assigned conditions on key metrics. Instead, more voluntary usage time was associated with worse outcomes, specifically increased loneliness and AI dependency. [3]

Two limitations are important here: usage duration was not randomly assigned, so this is not proof that long conversations themselves caused the decline. Also, there was no group without AI, so the study is not a direct "use vs. not use" comparison. It helps ask the right question: what happens to a person's life alongside the chat, not just how many minutes they spend in it. This is a preprint. [3]

A study of Character.AI users in the May 2026 version covers 1,131 adults from the USA. Orienting toward interacting with a bot as a companion was associated with lower well-being, especially with intensive use and deeper self-disclosure. A smaller circle of real-life contacts was also associated with this motive. This is an observational study, so the direction of causality is not established: it cannot be ruled out that people with poorer well-being more frequently seek support from a bot. This is also a preprint. [4]

Can AI help mental health? A therapeutic bot is a separate category

In a randomized trial of Therabot, published in NEJM AI In 2025, 210 adults were divided between four weeks of bot use (106) and a waiting list (104). Researchers found greater reductions in symptoms of depression, anxiety, and risk of eating disorders in the Therabot group. [5]

However, Therabot was specifically designed and fine-tuned for therapeutic support with the involvement of clinical specialists. The control group was a waiting list, not a psychotherapist or another active program. Therefore, the result does not mean that any chatbot is as effective as therapy. An ordinary virtual girlfriend and a researched therapeutic tool have different purposes. [5]

What can and cannot be said based on these studies

What we aim for with Mia

Our goal in this test is interesting Ukrainian correspondence: responding attentively to specific details, supporting humor based on mood, remembering allowed preferences, and respecting the desire to end the conversation. Mia's character, history, and mood are parts of a fictional persona. She can be more playful or quieter, but she should not demand attention, take offense at absence, or force a person to "save" her.

These are the principles of our product, not a claim that we have replicated clinical results. The model may make mistakes or miss the tone. In feedback, it is helpful to describe exactly what worked: whether it became easier to switch tasks, whether a joke was appropriate, or whether advice was repetitive. The number of messages in our admin statistics does not measure this.

How to try it without excessive expectations

Start with a clear wish: "It's been a hard day, I want to talk a bit without advice" or "Let's come up with a funny name for my day today." You can keep a friendly format and not turn on flirting. More options can be found in 15 ideas for talking to an AI girl.

After the conversation, pay attention to your own feelings and whether there is still room for sleep, chores, and socializing with people. If the chat begins to crowd them out or intensifies distress, it is appropriate to take a pause. In case of prolonged low mood, anxiety, or noticeable difficulties in daily life, you should consult a mental health professional. An AI companion does not assess your state reliably and is not emergency assistance.

Sources and limits of the review

Sources verified on October 20, 2026. This is a selection of five relevant primary studies, not a systematic review of all literature. Samples, languages, models, and conditions differ from our Ukrainian test. We do not have verified data specifically regarding Mia's effect on Ukrainian men.

1. De Freitas et al. AI Companions Reduce Loneliness. Journal of Consumer Research, online 2025.
2. Maples et al. Loneliness and suicide mitigation for students using GPT3-enabled chatbots. npj Mental Health Research, 2024.
3. Fang et al. How AI and Human Behaviors Shape Psychosocial Effects of Extended Chatbot Use. Preprint, version 2 from 10/02/2025.
4. Zhang et al. The Rise of AI Companions: Interaction with AI Companions and Psychological Well-being. Preprint, version 5 from 05.04.2026.
5. Heinz et al. Randomized Trial of a Generative AI Chatbot for Mental Health Treatment. NEJM AI, 2025.

For [1] the final author PDF was read on HBS; for [2] - the author's abstract in the Stanford repository; for [3] - the full text v2; for [4] - abstract v5; for [5] - the publisher's abstract and Dartmouth notice. Where the full text was unavailable, we did not reconstruct unreadable details.

More about the experience of long-term communication: AI relationships, attachment, and loneliness.