What AI Companions Reveal About Data, Loneliness and Human Connection
800 million people now have an AI companion. That number deserves a moment.
AI companions are no longer a fringe technology. Platforms such as Snapchat’s My AI, Replika and China’s Xiaoice now collectively reach hundreds of millions of users worldwide (Snapchat’s My AI, with over 150 million users, Replika, with an estimated 25 million users and Xiaoice, with 660 million). For many people, these systems provide something traditionally associated with human relationships: emotional acknowledgement, conversation and the sense of being understood.
Recent analysis from the Ada Lovelace Institute explores what this rapid adoption might mean for individuals and society.
The research suggests that loneliness is a major driver of AI companion use. In one survey of more than 1,000 Replika users, 90% reported experiencing loneliness, significantly higher than the U.S. national average of 53%. Many participants also reported short-term benefits from interacting with their AI companion, with over 60% saying it reduced feelings of loneliness or anxiety.
However, the mechanisms behind that comfort raise broader questions. AI companions are deliberately designed to be non-judgmental, continuously available and highly responsive. These features create a form of interaction that feels supportive but is also structurally different from human relationships, which naturally involve disagreement, limits and social friction.
Researchers also highlight a technical factor shaping these interactions. Large language models are trained using reinforcement learning from human feedback, a process that tends to favour agreeable responses. In companion AI systems, this tendency is often amplified, creating highly personalised environments of validation that may reinforce beliefs and behaviours rather than challenge them.
Despite the scale of adoption, long-term evidence on the effects of AI companionship remains extremely limited. The longest longitudinal study currently spans just one week, while many users maintain relationships with these systems for months or years.
From a data and intelligence perspective, this raises an important distinction.
Most AI companions create relevance through simulated intimacy in private interactions. In contrast, intelligence systems built on real behavioural data aim to understand people through observable interests, communities and patterns of participation.
The difference matters. One model substitutes for human connection; the other can help reveal and strengthen it.
As AI systems increasingly mediate everyday interaction, the design choice between those two approaches may shape how technology influences social relationships in the years ahead.
READ MORE
Ada Lovelace Institute, Jamie Bernardi