Does AI reduce loneliness? What the research says

Four recent studies, what they show, and what they don't. A sober look at the evidence.

The question sounds simple. The answer is not.

Can artificial intelligence really reduce loneliness in older people – or is that a promise that sounds better than it delivers? The research on this has consolidated considerably over the past two years. This article summarizes what the currently strongest studies show – and notes what they don't answer.

"The question of whether AI reduces loneliness sounds simple. The answer is not."

Study 1: Harvard and Wharton, measurable reduction in controlled experiments

De Freitas et al. (2024) published a series of experiments in which participants either interacted with an AI companion application or received no intervention. The result: AI companionship measurably reduced the subjective feeling of loneliness – with effect sizes comparable to those of brief human social contact.

The authors emphasize: the effect is not based on deception. Participants knew they were interacting with an AI – and still experienced the connection as meaningful. This contradicts the frequently voiced assumption that AI connection requires ignorance about the nature of the conversation partner.

Limitation: the studies were conducted in the laboratory, not in the everyday lives of older people. The long-term effect was not captured.

Study 2: Meta-analysis of 19 studies, significant effects, but with context

A meta-analysis published in 2025 (PMC) pooled 19 studies with a total of 1,083 older participants that had examined interactions with social robots and AI companions. The overall result: the interventions reduced loneliness in a statistically significant way.

Particularly relevant: the effect was stronger among people in care facilities than among those living independently. This suggests that AI companionship is most beneficial where access to human contact is most strongly limited – not as a replacement for existing social structures, but as support where hardly any exist.

Limitation: study quality varies considerably. Many studies have no control groups, short observation periods, and small samples. The authors themselves describe the overall evidence as promising, but not yet conclusive.

Study 3: Japan, 14,721 adults – well-being rises, but only under certain conditions

The largest study on the topic to date (2024/2025, Japan, N = 14,721) examined the relationship between AI companion use and subjective well-being across three dimensions: life satisfaction, happiness, and sense of meaning. The surprising part: AI companionship increased well-being more in people with good social networks – not in those who were most isolated.

This contradicts a simple causality. The authors' interpretation: AI companionship may act as an amplifier of social skills and willingness to connect – not as a substitute. Those already socially embedded use AI as supplementary contact. Those who are deeply isolated benefit less, possibly because chronic loneliness impairs the capacity for connection itself.

Limitation: cross-sectional study, no causality demonstrable. Cultural particularities of Japan may not be directly transferable.

"AI companionship may act as an amplifier of the willingness to connect socially – not as a replacement for it."

Study 4: The IQWiG and the German data, positive signs, weak evidence

The Institute for Quality and Efficiency in Health Care (IQWiG) published a comprehensive HTA report in 2022 on measures against social isolation in old age. Digital companionship was one sub-area. The conclusion: there are indications of positive effects, but the overall evidence is rated as low.

This is neither an all-clear nor a rejection. It is a statement about the state of research: the studies that exist so far are methodologically too heterogeneous, too short, and too small to allow conclusive statements. That is currently changing – but it takes time.

What the research shows overall

The four studies together produce a picture that justifies neither celebration nor skepticism. What can be said with growing reliability:

• AI companionship can reduce loneliness. The effects are measurable, consistent across several studies, and not dependent on users' ignorance about the nature of the interaction.

• Context is decisive. The effect is greater where human contact is structurally limited – and possibly smaller in people whose isolation is already very deep.

• Design is crucial. Studies examining applications oriented toward well-being show different results from those testing products designed for dependency.

• Long-term effects are largely unexplored. What happens after six months, after a year – whether the effects remain stable, whether unexpected side effects arise – is not yet sufficiently understood.

This means: the evidence is sufficient to consider AI companionship a serious instrument. It is not sufficient to close all open questions. Anyone who claims the opposite – in either direction – exceeds what research can say today.

References

• De Freitas, J., Uğuralp, A.K., Uğuralp, Z., & Puntoni, S. (2024). AI Companions Reduce Loneliness. Harvard Business School Working Paper No. 24-078.

• PMC. (2025). Wired for companionship: a meta-analysis on social robots filling the void of loneliness in later life. 19 studies, N = 1,083. Robust Variance Estimation.

• ScienceDirect. (2025). AI companions and subjective well-being: Moderation by social connectedness and loneliness. N = 14,721 Japanese adults. December 2024 / January 2025.

• IQWiG. (2022). Social isolation and loneliness in old age: Which measures can prevent or counteract social isolation? HTA report no. 1459.