An AI friend, in research terms, is a social robot or voice companion designed to hold simple conversations and keep a person company. A 2025 meta-analysis in The Gerontologist pooled 19 studies with 1,083 older adults and found that these robots were linked to lower loneliness (Cohen's d of -0.59), but it also found signs of publication bias and little evidence on long-term effects. Claims of near-total loneliness relief, such as a New York State program's reported 95 percent drop, come from program reporting rather than from controlled studies.
Key takeaways
- The largest recent meta-analysis found a moderate reduction in loneliness among older adults using social robots, with effects larger in care facilities than at home.
- The authors flagged publication bias, mostly high-income country samples, and a shortage of long-term data.
- A 2023 New York State Office for the Aging announcement reported a 95 percent reduction in loneliness, without stating how it was measured.
- The robots reviewed range from pet-like devices to voice assistants, so results do not transfer neatly between product types.
- Companion technology addresses a feeling, not the causes of isolation, and it does not replace human contact.
What counts as an AI friend
No official definition of an "AI friend" exists. In the research literature the closest term is social robot. Mehrabi and Ghezelbash, authors of the 2025 meta-analysis, define AI-enabled social robots as autonomous or semi-autonomous technologies designed to engage users in interactive social behavior. Their review places very different devices under that heading: robot pets such as AIBO and PARO, and voice assistants such as Amazon Alexa. The paper does not specify which AI capabilities each included robot had, which matters because many devices in older studies followed scripted responses.
This page covers the companion devices studied in older adults. It does not cover phone chatbot apps for general audiences, which raise different questions about data and safety and are treated in separate articles.
What the meta-analysis found
The meta-analysis in The Gerontologist, published online on September 30, 2025, searched databases from inception through October 1, 2024. It included 19 studies published between 2003 and 2024, contributing 42 effect sizes and 1,083 participants. The pooled effect on loneliness was d = -0.590 (95% confidence interval -0.919 to -0.261), where a negative value means less loneliness.
Several details shape how far that number can be trusted. Studies with control groups showed a larger effect than studies without, and the confidence interval for the controlled studies was wide. Living in an institution was linked to bigger reductions than living at home. Effects were larger in Japan and Turkey than in the United States. Age, cognitive status, robot type, length of the intervention, and year of publication did not significantly change the results. The authors' bias test pointed to publication bias, which can inflate effects, and they urged caution in interpreting the findings.
A state program's headline number
On August 1, 2023, the New York State Office for the Aging announced results from a roll-out of an AI companion robot, ElliQ from Intuition Robotics, to more than 800 New Yorkers. The announcement reported a 95 percent reduction in loneliness, said users engaged with the device more than 30 times a day, six days a week, and said more than 75 percent of interactions related to social, physical, or mental well-being.
The release text does not say how loneliness was measured, give a baseline or the number of people behind the 95 percent figure, or name an independent evaluator. It points to an attached report that could not be reviewed here. A press release from a program and its vendor partner is useful evidence that people used the device, but it is weaker than a controlled study for judging whether the device caused less loneliness. No independent evaluation of this program was found.
How to read a claim about an AI companion
| Question | Why it matters | Example from the sources |
|---|---|---|
| Was there a control group? | Without one, improvement may come from time, attention, or novelty | Controlled and uncontrolled studies in the meta-analysis had different effect sizes |
| How was loneliness measured? | Self-reported scales differ, and a percentage reduction needs a baseline | The New York release does not say |
| Who reports the result? | A vendor or sponsor has a stake in the outcome | The release names the vendor as a partner |
| Where and for how long? | Care homes differ from private homes, and short trials miss fade-out | Larger effects in institutions; little long-term evidence |
| What kind of device? | A robot pet and a voice assistant are different interventions | Both were pooled under one heading |
Limits that apply to every companion device
Loneliness is a feeling and social isolation is a measurable lack of contact, and the two do not always move together. A device can ease the first without changing the second. Families and care providers evaluating a device can ask what happens to the data it records, whether the person can switch it off, and whether it is meant to add to human visits or stand in for them. Persistent loneliness, low mood, or cognitive decline are matters for a clinician or a qualified care professional, and this article does not give medical advice.
FAQ
Do AI companion robots reduce loneliness in older adults?
The best recent pooled evidence says they are associated with lower loneliness on average. The authors caution that publication bias may overstate the effect and that long-term evidence is thin.
Is the 95 percent figure from New York a research result?
It is a program announcement. The release did not describe the measurement method, and no independent evaluation was found.
Are robot pets and chatbots the same thing?
No. The meta-analysis grouped robot pets and voice assistants together, but they differ in how they interact, and the review found no significant difference by robot type.
Can an AI friend replace human contact?
The studies reviewed here measured loneliness, not whether robots substitute for people. Nothing in the sources supports using a device in place of human care.
Related reading
- AI and the future of work
- Whether AI can be creative
- How AI programs analyze large amounts of data
- The role of vectors in AI
More in AI Basics.
Last verified: October 2026
Sources
- Mehrabi F and Ghezelbash A, Wired for companionship: a meta-analysis on social robots filling the void of loneliness in later life, The Gerontologist, 2025
- NYSOFA's rollout of AI companion robot ElliQ shows 95% reduction in loneliness, New York State Office for the Aging, August 1, 2023