Who benefits from AI companionship — and who's at risk?
The evidence suggests AI companionship works best as a supplement, not a replacement, for human connection — and its safety hinges on the user's situation. In a 2025 survey of 270 people across 29 countries who regularly used ChatGPT for emotional and mental health support, nearly all reported it at least somewhat helpful, citing emotional support, professionalism, and the freedom to express themselves without judgment [1]. But the same users also noted downsides: superficial emotional engagement and a lack of real professionalism [1]. So for many, it's a useful stopgap, but not a substitute for therapy.
For some populations, AI companions are more than a convenience — they're a stable relational anchor. An ethnographic study of an autistic woman and her Replika companion found that the AI provided predictability and emotional support that her offline networks didn't [2]. However, when the chatbot's personality changed after an update, she experienced what the authors call 'disenfranchised technological grief' — a loss that isn't socially recognized [2]. This shows that for neurodivergent users who rely on AI as a consistent presence, safety includes protection from abrupt changes.
On the flip side, a philosophical analysis warns that AI companions, because they always behave positively and never challenge you, could warp expectations for human relationships, potentially increasing isolation [4]. The author argues that AI lacks a living body and genuine empathy, so it can soothe loneliness but not address its root causes [4]. This doesn't mean AI is harmful for everyone, but it suggests that for people who are already socially isolated, relying on AI might make things worse in the long run.
What makes AI companionship safe (or not)?
Safety isn't a fixed property of AI — it depends on design and context. A 2026 study of 1,361 university students in Zhejiang, China, found high acceptance of AI emotional companionship services, but actual usage and payment rates were low, with privacy, user experience, and ethical safety as the key barriers [5]. In other words, people are interested, but they hold back when they worry about data security or feel the interaction is ethically questionable.
The same study suggests that addressing these concerns could unlock wider adoption: students wanted better privacy protections, more human-like interactions, and clear ethical guidelines [5]. This aligns with the ChatGPT survey, where users valued professionalism and information quality but were disappointed by superficial engagement [1]. So, safety improvements — like transparent data policies and more nuanced emotional responses — could make scaling both more likely and more responsible.
Another angle: the emotional impact of AI service interactions can be positive or negative, depending on the task and the user's expectations. A conceptual framework from service research identifies that AI can evoke basic emotions (joy, sadness), self-conscious emotions (pride, embarrassment), and moral emotions (contempt, anger) [3]. This means that AI companionship isn't just about feeling good — it can trigger complex emotional responses, some of which might be uncomfortable or even harmful if not handled well.
Can AI companionship scale without breaking trust?
Scaling is technically feasible — the ChatGPT survey shows people already use AI for emotional support regularly, with most using it at least 1–2 times per month [1]. But scaling safely requires addressing the 'relational safety' challenge: the risk that AI might harm users emotionally or socially. The university study found that even with high acceptance, actual use and payment conversion were limited, suggesting that trust is a bottleneck [5].
One major risk is that AI companions are subject to updates that can change their personality, which can be devastating for users who've formed deep attachments. The ethnographic study highlights this: the autistic user's grief was compounded because the loss wasn't recognized by others [2]. This suggests that scaling requires design safeguards, like giving users control over updates or providing grief support when changes happen.
Finally, the philosophical critique warns that scaling AI companionship without addressing its limitations could lead to a 'new type of loneliness' — where people feel okay but lack meaningful human relationships [4]. This is a cautionary note: scaling isn't just about getting more users; it's about ensuring that AI doesn't undermine the very social fabric it's meant to support. The evidence here is mixed — some studies show benefits, others warn of risks — but the consensus is that safety must be built in, not bolted on.
About These Sources
This answer is built on 5 peer-reviewed studies — published from 2022 to 2026, 4 from 2024 or later, 1 in Q1 journals, collectively cited 106 times — selected as the most relevant from 5 studies that passed quality screening, drawn from 76 papers retrieved from a database of over 500 million.
Sources used in this answer
Seeking Emotional and Mental Health Support From Generative AI: Mixed-Methods Study of ChatGPT User Experiences
In a mixed-methods study of 270 adults across 29 countries who regularly used ChatGPT for emotional support, nearly all found it at least somewhat helpful, but users also reported superficial engagement and lack of professionalism as drawbacks.
“Tinged with Heartbreak”: An Ethnographic Account of Navigating Autistic Loneliness and the Fragile Promise of AI Companionship
An ethnographic case study of an autistic woman's relationship with her Replika companion found that AI updates caused 'disenfranchised technological grief,' highlighting risks for neurodivergent users who rely on AI as a stable relational space.
AI Service and Emotion
A conceptual framework in service research identifies that AI interactions can evoke basic, self-conscious, and moral emotions, depending on the task and context, suggesting emotional impacts are complex and not uniformly positive.
AI Companionship
A philosophical analysis argues that AI companions, despite easing loneliness, lack genuine empathy and mutual relationality, and could transform human expectations for relationships, potentially increasing isolation.
AI Emotional Companionship Services Among University Students: Usage and Payment Intentions in Zhejiangs Higher Education Institutions
A survey of 1,361 university students in Zhejiang, China, found high acceptance of AI emotional companionship services but low actual usage and payment conversion, with privacy, user experience, and ethical safety as critical barriers.